Draft Technical Report
CONSIDERATIONS RELATED TO POST CLOSURE MONITORING OF
URANIUM IN-SITU LEACH/IN-SITU RECOVERY (ISL/ISR) SITES
Background Information Document for the Revision of 40 CFR Part 192
Radiation Protection Division
Office of Air and Radiation
U.S. Environmental Protection Agency
Revision 8
September 2014
EPA-402-D-14-001
-------
TABLE OF CONTENTS
Acronyms and Abbreviations ix
Overview 1
Background 1
Overview of Report Contents 2
1.0 Introduction 4
1.1 Background versus Baseline Wells 7
1.2 ISR Facilities in the United States 9
2.0 Resource Conservation and Recovery Act 14
2.1 Summary 14
2.2 Ground Water Monitoring Requirements for Treatment, Storage, and
Disposal Facilities 15
2.2.1 Overview 15
2.2.2 Permitted Facilities 15
2.2.3 Detection Monitoring 16
2.2.4 Compliance Monitoring 17
2.2.5 Corrective Action 17
2.3 Application to ISR Facilities 18
3.0 Ground Water Monitoring at ISR Facilities 20
3.1 Overview 20
3.2 Pre-operational Monitoring (Phase 1) 21
3.3 The ISR Leaching Process (Phase 2) 22
3.3.1 Excursions during Operations 24
3.4 Post-operational Monitoring (Phases 3 through 5) 30
3.5 Selection of Parameters to Be Used in Ground Water Sampling Programs 31
3.5.1 Regulated Constituents 31
3.5.2 Summary of Species Potentially Required for Compliance
Monitoring- Tiered Approach 39
3.5.3 Well Construction and Low-Flow Sampling Methodologies 42
3.5.4 Species Required for Geochemical Modeling 45
3.5.5 Species Required for Excursion Monitoring 48
3.5.6 Case History -Evolution of Constituent Monitoring List 49
3.5.7 Formal Approach to Acceptable Restoration 50
4.0 Technical Considerations for ISR Ground Water Monitoring 54
4.1 Uranium Geology 54
4.1.1 Formation of Uranium Containing Ore Deposits 55
4.2 Aquifer Exemption Requirement 59
4.3 Establishing Baseline Conditions 59
4.3.1 Variability in Baseline Measurements 64
4.4 Extraction Operations Phase 68
4.5 Post-extraction Phase 70
4.6 Factors Affecting Post-mining Time Frames and Wellfield Stability 70
4.7 Modeling 73
Draft Technical Report ii Revised Draft-September 2014
-------
4.7.1 Objectives and Conceptual Model Development 73
4.7.2 Ground Water Flow and Contaminant Transport Modeling 75
4.7.3 Geochemical Modeling 77
4.7.4 Demonstrating Long-term Stability of Restored ISR Wellfields -
Long-Term Monitoring and Geochemical Modeling 91
5.0 Active/Existing ISR Facilities: Monitoring Issues 94
5.1 Ground Water Baseline: Case Studies 94
5.2 Wellfield Restoration 96
5.3 Wellfield Restoration: Case Study 97
6.0 Issues Associated with Establishment of Post-restoration Steady State 99
6.1 Post-restoration Stability Monitoring 99
6.2 Factors That Affect Time Frames for Post-mining Monitoring 100
6.2.1 Fate and Transport Processes 100
6.2.2 Natural Attenuation Processes 103
6.3 Geochemically Based Restoration Techniques 109
6.4 Monitored Natural Attenuation 110
6.4.1 Tiered Approach to Assessing Suitability of Monitored Natural
Attenuation 112
6.4.2 First-Order Attenuation Rate Determination 112
7.0 Statistical Analyses to Compare Pre- and Post-ISR Conditions 115
7.1 Determine Baseline Characteristics 119
7.1.1 Design for Baseline Sampling 122
7.1.2 Selection of Baseline Monitoring Wells 123
7.1.3 Determining the Number of Baseline Samples 126
7.1.4 Summary 130
7.2 Determining the Number of Monitoring Wells Required to Detect
Noncompliance 136
7.2.1 Determining the Number of Monitoring Wells based on
Hypergeometric Sampling 136
7.2.2 An Alternative Graduated Approach to Hypergeometric Sampling 146
7.2.3 Determining Connectivity of the Wellfield 160
7.3 Hypothesis Testing and Data Quality Objectives 162
7.3.1 Decision Errors and Confidence Levels 164
7.3.2 Hypothesis Tests for Comparisons with Baseline 165
7.3.3 Selecting a Test Form 168
7.3.4 Hypothesis Tests for Detecting Trends 170
7.4 Selecting the Statistical Approach - Parametric Versus Nonparametric
Methods 172
7.4.1 Determining If Data Have a Normal Distribution 173
7.4.2 The Shapiro-Wilk W Test 173
7.4.3 The Studentized Range Test 173
7.5 Outlier Detection 174
7.5.1 Parametric Tolerance Limits for Outliers 175
7.5.2 Calculating an Upper Tolerance Limit 176
7.6 Determining the Number of Samples per Well 178
Draft Technical Report iii Revised Draft-September 2014
-------
7.6.1 Same Sample Sizes (n=m) and Same Standard Deviation (01=02) 179
7.6.2 Different Sample Sizes (n^m), Same Standard Deviations (01=02) 180
7.6.3 Different Sample Sizes (n^rn) and Different Standard Deviations
(o^o2) 180
7.7 Statistical Methods for Trends and Seasonality 181
7.7.1 Adjusting for Seasonality 182
7.7.2 Using Trend Tests to Determine Stability 184
7.8 Analysis of Post-restoration Trends at ISR Sites 199
7.8.1 Trend Analysis by Well 199
7.8.2 Pooled Trend Analysis 210
7.9 Verify that Contaminants and Hazardous Constituent Concentrations are
Below Required Restoration Levels 214
7.9.1 Parametric Method for Determining Compliance of Individual Wells ...215
7.9.2 Nonparametric Tests for Comparing Baseline and Post-restoration
Conditions 220
7.10 ProUCL Software for Statistical Analysis 223
7.11 Summary of Statistical Approaches 224
8.0 Summary of Post-closure Performance Issues 226
8.1 Designing the Monitoring Program to Allow Reliable Baseline Conditions
to be Established Prior to Active Mining 226
8.2 Determining that the Ground Water Chemistry has Reached Steady State
and Restoration Processes can be Discontinued 228
8.3 Post-restoration Stability Monitoring 229
9.0 References 232
ATTACHMENTS
Attachment A: Development of Ground Water Baseline for Dewey-Burdock ISL/ISR Site in
South Dakota
Attachment B: Post-restoration Stability Monitoring Case Histories
Attachment C: Aquifer Restoration (Extracted from NRC 2009, Section 2.11.5)
Attachment D: Instructions and Examples for Statistical Calculations
Attachment E: Statistical Tables
Attachment F: Detailed Results of Regression Trend Analysis by Analyte and Mine Unit
Attachment G: Using Trend Tests to Determine Stability
Attachment H: Glossary of Terms
Draft Technical Report iv Revised Draft-September 2014
-------
LIST OF TABLES
Table 1-1. NRC-Licensed ISR Facilities as of September 2011 9
Table 1-2. Uranium Recovery Facility Applications, Reviews, and Letters of Intent by
NRC 11
Table 1-3. List of In-Situ Uranium Mines in Texas 13
Table 3-1. Wells on Excursion at Christensen Ranch - 2Q 2011 30
Table 3-2. Ground Water Species Identified in EPA Regulations That May
Require Monitoring at ISR Facilities 35
Table 3-3. Comparison of Ground Water Parameter Measurements Established by
Various Regulators with Actual Field Measurements from Dewey-Burdock
Site 39
Table 3-4. Parameters for Solid Phase Characterization 47
Table 3-5. Basis for Inclusion of Various Analytes in ISR
Ground Water Monitoring Program 51
Table 4-1. Summary of Commonly Applied Geochemical Modeling Codes 86
Table 5-1. Baseline Water Quality Data for ZamzowPAA-1 95
Table 5-2. Ground Water Chemistry of Texas In-Situ Uranium Production
Authorization Areas 98
Table 6-1. Post-restoration and Stability Monitoring Periods 100
Table 7-1. Outline of the Statistical Procedures used in Phases 1, 4, and 5 117
Table 7-2. Summary Statistics for Population Coefficient of Variation (cv) of Baseline
Parameters at Nine ISR Sites 130
Table 7-3. Baseline Statistics and Number of Samples Required at Christensen Mine
Unit 6 for the Relative Standard Error of the Baseline Mean to Be Less
Than P% for 35 Analytes with Summary Statistics for U and Ra-226 132
Table 7-4. Wellfield Characteristics and Comparison of Actual and Target Baseline
Sample Size at Nine ISR Production Units 134
Table 7-5. Number of Samples Required at Nine Production Units for Relative
Standard Error of U and Ra-226 Mean Baseline Concentrations to Be Less
Than±P% 135
Table 7-6. Parameter Definitions for the Hypergeometric Distribution 139
Table 7-7. Minimum Value of N with Prob {Q < q x=0, M, N} > 0.95 142
Table 7-8. Minimum Value of N with Prob {Q < q x=l,M, N} > 0.95 144
Table 7-9. Ratio of Monitor Wells to Production Wells 156
Table 7-10. Number of Monitoring Wells Required for Five Design Options for a
Production Unit with 181 Wells in 5-spot Pattern (Posterior
Probability=90%) 156
Draft Technical Report
Revised Draft - September 2014
-------
Table 7-11. Number of Monitoring Wells Required for Five Design Options for a
Production Unit with 181 Wells in 5-spot Pattern (Posterior
Probability=95%) 156
Table 7-12. Number of Monitoring Wells Required for Four Design Options for a
Production Unit with 162 Wells in 7-spot Pattern (Posterior
Probability=90%) 157
Table 7-13. Number of Monitoring Wells Required for Four Design Options for a
Production Unit with 162 Wells in 7-spot Pattern (Posterior
Probability=95%) 157
Table 7-14. Number of Monitoring Wells Required for Posterior Probability of 90% that
at Least P% of the Well Zones Demonstrate Compliance 158
Table 7-15. Number of Monitoring Wells Required for Posterior Probability of 95% that
at Least P% of the Well Zones Demonstrate Compliance 159
Table 7-16. Hypothesis Testing: Type I and Type II Errors 164
Table 7-17. Critical Values for the Studentized Range Test 174
Table 7-18. One-Sided Upper Tolerance Limit Factors with g% Coverage
for Selected Values of N 177
Table 7-19. Number of Quarterly Samples Required for 90% Probability of Detecting
Slope Using a Mann-Kendall or Regression Trend Test 187
Table 7-20. Number of Quarterly Samples Required for 95% Probability of Detecting
Slope Using a Mann-Kendall or Regression Trend Test 188
Table 7-21. Number of Quarterly Samples Required for 99% Probability of Detecting
Slope Using a Mann-Kendall or Regression Trend Test 189
Table 7-22. Probability of Detecting a Trend with 12 Quarterly Samples 191
Table 7-23. Probability of Detecting a Trend with 20 Quarterly Samples 192
Table 7-24. Probability of Detecting a Trend with 32 Quarterly Samples 193
Table 7-25. Number of Samples and Number of Years Required for 95% Chance of
Detection Using Regression or Mann-Kendall Test 195
Table 7-26. Regression Statistics for Example in Figure 7-16 200
Table 7-27. Summary of Trend Analysis at Four* ISR Sites 206
Table 7-28. Mean Slope and Variability Estimates with 95% Confidence Interval for the
Mean (LCL to UCL) 207
Table 7-29. Summary of Significant Trends in Pooled Trend Analysis 214
Table 7-30. Significant Positive and Negative Trends Identified using Pooled Trend
Analysis 214
Table 7-31. Critical Values of the Student's ^-Distribution 217
Draft Technical Report
VI
Revised Draft - September 2014
-------
LIST OF FIGURES
Figure 1-1. Historical U.S. Uranium Production after World War II 5
Figure 1-2. In-situ Uranium Recovery -Process Flow Diagram 6
Figure 3-1. Variation of Typical Ground Water Constituent over Time 21
Figure 3-2. Idealized Schematic Cross Section to Illustrate Ore-Zone Geology and
Lixiviant Migration from an Injection Well to a Production Well (NRC
2009) 22
Figure 3-3. Schematic Diagram of a Wellfield Showing Typical Injection/Production
Well Patterns, Monitoring Wells, Manifold Buildings, and Pipelines (NRC
2009) 23
Figure 3-4. Wellheads and Header House, Smith Ranch, Converse County, Wyoming 24
Figure 3-5. Alkalinity (mg/L) Variation during Excursion of CrowButte 28
Figure 3-6. Conductivity (|imho/cm) Variation during Excursion of Crow Butte 28
Figure 3-7. Chloride (mg/L) Variation during Excursion of CrowButte 29
Figure 3-8. Cross Section of a Typical Injection, Production, or Monitoring Well 44
Figure 3-9. Major Chemical Reactions Involved in Uranium Recovery and Restoration 46
Figure 4-1. Uranium Resource Areas of the United States 54
Figure 4-2. Pending, Licensed, and Active ISL Operations 55
Figure 4-3. Three-Dimensional Depiction of Uranium Ore Deposited in Paleochannels 56
Figure 4-4. Schematic Diagrams of the Different Geometries for Tabular, Roll-front,
Fault Displaced, and Remnant Ore 56
Figure 4-5. Conceptual Cross-Section of Uranium Roll-front Deposits 57
Figure 4-6. Conceptual Model of Uranium Roll-front Deposit 58
Figure 4-7. Well and Production Zone Locations and Baseline Concentrations of TDS,
Uranium, and Radium -Wellfield H-E 63
Figure 4-8. Baseline Uranium Concentrations at the Rosita ISR Facility 64
Figure 4-9. Schematic Diagram of a Wellfield Showing Typical Injection/Production
Well Patterns, Monitoring Wells 68
Figure 4-10. Example of MODFLOW Predicted Potentiometric Surface during Active
Mining 69
Figure 4-11. Example of MODPATH Predicted Flow Paths During Active Mining 69
Figure 4-12. Example of PHT3D Predicted Post-mining Uraninite Concentrations 71
Figure 7-1. Ore Zone Outline and Well Locations at Christensen Ranch Mine Unit 6 128
Figure 7-2. Scatter Plot of the Maximum RSEM for U and Ra-226 Mean Concentrations
versus Baseline Sample Size at Nine Production Units 136
Figure 7-3. 2-Well Design for Production Unit with Minimal Connectivity 149
Draft Technical Report vii Revised Draft-September 2014
-------
Figure 7-4. 5-Well Design for Production Unit with Low Connectivity 149
Figure 7-5. 7-Well Design for Production Unit with Moderate Connectivity 150
Figure 7-6. 11 -Well Design for Production Unit with Good Connectivity 150
Figure 7-7. 20-Well Design for Production Unit with High Connectivity 151
Figure 7-8. 2-Well Design for Production Unit with Minimal Connectivity 152
Figure 7-9. 4-Well Design for Production Unit with Low Connectivity 153
Figure 7-10. 10-Well Design for Production Unit with Good Connectivity 154
Figure 7-11. 18-Well Design for Production Unit with High Connectivity 155
Figure 7-12. Test Performance Plot with Parameter Definitions for Test Form 1 167
Figure 7-13. Test Performance Plot with Parameter Definitions for Test Form 2 168
Figure 7-14. Plot of Number of Samples and Number of Quarters Required for 95%
Chance of Detection Using Regression versus Sampling Frequency 196
Figure 7-15. Uranium Concentrations in Crow Butte Well PR-15 202
Figure 7-16. Deviation of Uranium Concentration from Mean with
Variability Bounds (±lo) 202
Figure 7-17. Radium Concentrations in Crow Butte Well PR-15 203
Figure 7-18. Comparison of Bayesian Posterior Distributions for Regression Slope
Parameter for Uranium and Radium in Crow Butte Well PR-15 203
Figure 7-19. Comparison of Complementary Cumulative Distribution Functions (CCDF)
for Regression Slope Parameter for Uranium and Radium in Crow Butte
WellPR-15 204
Figure 7-20. Comparison of Bayesian Posterior Distributions for Regression Slope
Parameter for Radium in Crow Butte Wells PR-Sand PR-15 204
Figure 7-21. Slope of Trend Line Averaged over All Wells 207
Figure 7-22. Temporal Variability (Averaged Over All Wells) 208
Figure 7-23. 95% Confidence Interval for Mean Temporal Variability 208
Figure 7-24. Full Range of Temporal Variability 209
Figure 7-25. Scatter Plot of the Standard Error versus the Number of Samples 209
Figure 7-26. Christensen MU2 Chloride Samples over Time with Trend Line 211
Figure 7-27. Christensen MU2 Chloride Samples over Time with Trend Line 212
Figure 7-28. Christensen MU2 TDS Samples over Time with Trend Line 212
Figure 7-29. Scatter Plot of Standard Error of Slope versus Pooled Number of Samples 213
Figure 7-30. Scatter Plot of Pooled f-test Results 213
Draft Technical Report viii Revised Draft-September 2014
-------
ACRONYMS AND ABBREVIATIONS
ACL alternate concentration limit
ADAMS Agency-wide Documents Access and Management System
AFCEE Air Force Center for Environmental Excellence
API American Petroleum Institute
ASTM American Society for Testing and Materials
C Celsius
C vs. D concentration vs. distance
C vs. T concentration vs. time
CA component activity
CBR Crow Butte Resources
CCDF complementary cumulative distribution function
CCL Contaminant Candidate List
CDF cumulative distribution function
CERCLA Comprehensive Environmental Response, Compensation, and Liability Act
CES cost effective sampling
CFR Code of Federal Regulations
cm/sec centimeters per second
COGEMA COGEMA Mining, Inc.
cv coeffi ci ent of vari ati on
DOE Department of Energy (U. S.)
DQO data quality objective
EC electrical conductivity
EDTA ethylenediaminetetraacetic acid
Eh oxidation-reduction potential
EM excursion monitor wells
EMP production zone wells
EPA Environmental Protection Agency (U.S.)
ft feet
GC generalized composite
gpm gallons per minute
GTS geostatistical temporal/spatial
Draft Technical Report
IX
Revised Draft - September 2014
-------
GUI graphical user interface
GWPS ground water protection standard
HFO hydrous ferric oxide
H0 null hypothesis
HA alternative hypothesis
HPD highest posterior density
IMWA International Mine Water Association
ISL in-situ leaching
ISR in-situ recovery
ISWS Illinois State Water System
K hydraulic conductivity
Kd partition or distribution coefficient
LCL lower control limit
LTM long term monitoring
MAROS Monitoring and Remediation Optimization System
MARS SIM Multi-Agency Radiation Survey and Site Investigation Manual
MCL maximum contaminant level
MCLG maximum contaminant level goal
MDD minimum detectable difference
mg/L milligram per liter
MNA monitored natural attenuation
MRDL maximum residual disinfectant level
mrem/yr millirem per year
MU mine unit
NDEQ Nebraska Department of Environmental Quality
NFESC Naval Facilities Engineering Service Center
NMA National Mining Association
NPDWR National Primary Drinking Water Regulation
NRC Nuclear Regulatory Commission (U.S.)
NRMRL National Risk Management Research Laboratory
NSDWR National Secondary Drinking Water Regulation
NUREG U. S. Nuclear Regulation Commission Regulation
OSWER Office of Solid Waste and Emergency Response (EPA)
Draft Technical Report
Revised Draft - September 2014
-------
PAA Production Authorization Area
pCi/L picocurie per liter
PCL protective concentration level
pH measure of acidity of a solution
PL prediction limit
ppb parts per billion
PVC polyvinyl chloride
QAPP Quality Assurance Proj ect Plan
RAI request for additional information
RARE Regional Applied Research Effort
RCRA Resource Conservation Recovery Act
RDP restoration data package
RO reverse osmosis
RSEM relative standard error of the mean
SCL single indicator control limit
SCM Surface Complexation Model
SDWA Safe Drinking Water Act
SDWR Secondary Drinking Water Regulations
SI saturation index
SSI statistically significant increase
TCEQ Texas Commission on Environmental Quality
TDS total dissolved solids
TSDF treatment, storage, and disposal facility
UCL upper control limit
UIC underground injection control
URI Uranium Resources, Inc.
UMTRCA Uranium Mill Tailings Radiation Control Act
USDW underground source of drinking water
USGS U.S. Geological Survey
UTL upper tolerance limit
WDEQ Wyoming Department of Environmental Quality
WQD Water Quality Division (WDEQ)
WRS Wilcoxon Rank Sum
Draft Technical Report
XI
Revised Draft - September 2014
-------
a alpha
P beta
|j,mho micromho
Draft Technical Report xii Revised Draft-September 2014
-------
OVERVIEW
BACKGROUND
In accordance with the Uranium Mill Tailings Radiation Control Act of 1978 (UMTRCA),
section 206, the U.S. Environmental Protection Agency (EPA) is authorized to develop standards
for the protection of public health, safety, and the environment from radiological and
nonradiological hazards associated with residual radioactive materials at inactive uranium mill
tailings sites. The legislation also authorizes EPA to set standards for these hazards when they
are associated with the processing, possession, transfer, and disposal of byproduct material
(tailings or wastes) at sites where ores are processed primarily for their uranium content or used
for disposal of byproduct or residual radioactive materials. UMTRCA requires EPA to develop
health and environmental standards for both Title I inactive uranium milling sites administered
by the U.S. Department of Energy (DOE) and Title II operations licensed by the U.S. Nuclear
Regulatory Commission (NRC) or its Agreement States.
In 1983, EPA promulgated regulations at 40 CFR Part 192, "Health and Environmental
Protection Standards for Uranium and Thorium Mill Tailings," in response to the statutory
requirements of UMTRCA. When the Agency promulgated 40 CFR Part 192, uranium recovery
from ore was based almost exclusively on the conventional milling process. This process
recovered a few pounds of uranium for each ton of ore mined and processed. The residues from
the milling process (the tailings or byproduct material) accumulated in large piles on the surface
at the milling site. Concern that these tailings piles would be a continuing source of radiation
exposure unless properly reclaimed was the driving force behind the passage of UMTRCA.
Because the major environmental risk at that time was perceived to come from the conventional
uranium mill tailings, other uranium recovery operations, such as heap leaching and in-situ
leaching (ISL), received little attention.
EPA last revised its regulations for uranium and thorium milling in 1995, and currently is
reviewing them to determine if they need to be updated. Since 40 CFR Part 192 was
promulgated, uranium recovery has shifted from conventional milling to ISL where, in a sense, a
portion of the milling process is conducted underground. Where the ore body is amenable to use
of the ISL technology, uranium can be recovered economically without the extensive surface
facilities, large waste volumes, or expectations of long-term site maintenance associated with
conventional milling. In the ISL process, also referred to as in-situ recovery (ISR),1 chemical
solutions are pumped underground through an array of wells into the ore body, where the
uranium is dissolved in place. The uranium-rich solutions are pumped to the surface, where the
uranium is extracted. The solutions are then chemically refortified and pumped back into the ore
body to recover additional uranium.
EPA's standards must address nonradiological, as well as radiological, constituents. Therefore,
for Title I sites, UMTRCA states that the standards shall, "... to the maximum extent practicable,
1 The term in-situ recovery seems to be gaining more traction in regulatory and technical documents than
in-situ leaching, and therefore, through the balance of this document, in-situ recovery or ISR will be used. We note,
however, that in-situ leaching is a more precise description, since the leaching occurs in-situ (underground), but the
recovery of uranium occurs in surface facilities.
Draft Technical Report 1 Revised Draft-September 2014
-------
be consistent with the requirements of the Solid Waste Disposal Act, as amended," now known
as the Resource Conservation and Recovery Act (RCRA). For Title II sites, the nonradiological
standards shall be "... consistent with the standards required under Subtitle C of the Solid Waste
Disposal Act, as amended, which are applicable to such hazards."
EPA's current standards in 40 CFR Part 192 incorporate the RCRA ground water monitoring
requirements for hazardous waste units specified in 40 CFR Part 264, including statistical
techniques for use in determining when monitoring requirements have been achieved. A key
question in amending 40 CFR Part 192 is whether, and to what extent, it is appropriate to apply
these technical approaches (developed to address releases to ground water from engineered units
such as landfills, impoundments, and tanks) to ISR facilities, where the regulated "unit" is a
defined portion of an aquifer. The focus of EPA's current revision effort for Part 192 focuses on
the development of standards for ISR operations.
OVERVIEW OF REPORT CONTENTS
With ISR operations expected to be the most common type of new uranium extraction facility in
the United States, and the potential for these facilities to affect ground water, EPA is considering
how to address ground water monitoring as a component of the regulatory standards specifically
applicable to these facilities in its revision of 40 CFR Part 192. This report is designed to serve
as a background information document and to provide a technical discussion of relevant issues to
assist EPA in addressing revision of 40 CFR Part 192 to reflect ISR operations.
Monitoring an ISR uranium extraction operation has several objectives: to establish baseline
(pre-operational) ground water chemical compositions in the ore zone; to detect excursions of the
injected and mobilized components beyond the wellfield; and to determine when the post-
operational (restoration phase) ground water chemistry has "stabilized" (i.e., reached
concentration levels that are expected to remain constant over time). The focus of this report is
on monitoring to establish post-operational stability rather than on operational excursion
monitoring.
EPA has stated that the regulatory effort will focus on establishing requirements applicable to
ISR facilities. Because the "milling" of uranium ore is performed within the aquifer by injection
of mobilizing agents, ISR facilities present challenges for ground water protection that are
distinct from those posed by conventional mills. Furthermore, the intent of ISR operators is to
release the site for other uses after additional processing of ore is no longer economically viable.
Given the disruption of the aquifer inherent in ISR technology and the foreseeable desire for a
relatively short period of post-operational institutional control, ground water protection will be of
central importance in amendments to 40 CFR Part 192.
As noted above, one purpose of monitoring is to demonstrate that the aquifer conditions
(i.e., contaminant concentrations or geochemical characteristics) established at the end of
restoration are sustainable, or stable, overtime. Currently, the duration of stability monitoring is
a site-specific period of time established in the license(s) required by NRC or the appropriate
Agreement State. In the past, the license-established restoration period generally has been about
6 months. More recently, the trend has been to increase the monitoring period established in the
license to at least 1 year. In practice, the actual period necessary for contaminant concentrations
Draft Technical Report 2 Revised Draft-September 2014
-------
to stabilize may be several years, with iterative analyses of additional samples required by the
regulators.
This technical report is intended to support consideration of issues associated with establishing
the ground water baseline for new facilities, demonstrating that the restored ground water has
reached steady state, and showing that post-restoration stability monitoring ensures that the
ground water quality is not deteriorating over time after restoration.
This report provides a summary of UMTRCA, a summary of relevant components of RCRA,
background information on the ISR process, discussion of the purposes of a ground water
monitoring system, description of factors affecting the time frame and ability to restore an ISR
wellfield to baseline conditions; and discussion of various statistical techniques and approaches
to measure the achievement of post-operational restoration goals. The report includes case
studies, identifies key issues associated with post-closure monitoring, and summarizes
performance issues regarding ground water monitoring at ISR facilities.
The report is intended to provide the scientific support to provisions in the rulemaking,
particularly in the areas of chemical characterization of ground water in the affected areas under
pre- and post-mining conditions, statistical analyses of field data from both initial
characterization efforts prior to mining and analyses of post-restoration monitoring data, and
performance measures applied to the analyses of these data. Specific statistical methods are not
recommended for mandatory use at all ISR facilities. Rather, the choice of statistical techniques
should be based on the quantity and quality of the field data available for any specific site. The
broader regulatory requirements established in the rulemaking should guide ISR operators and
regulators to design field data collection activities to develop robust databases to support the use
of the statistical techniques used to measure the long-term performance of a restored ISR ore
zone.
Draft Technical Report 3 Revised Draft-September 2014
-------
1.0 INTRODUCTION
The U.S uranium mining industry has been highly cyclical over the past 60 years, a
phenomenon! typical of the mining industry in general. Until the late 1970s, uranium production
was based on conventional milling processes that involved leaching the mined ore to recover the
uranium values. The conventional milling processes resulted in large amonts of "tailings," the
by-product residues from leaching the ore. To address hazards associated with these mill tailings,
the U.S. Environmental Protection Agency (EPA) promulgated in 1983 regulations at
40 CFR Part 192, "Health and Environmental Protection Standards for Uranium and Thorium
Mill Tailings," in response to the statutory requirements of the Uranium Mill Tailings Radiation
Control Act (UMTRCA) of 1978. UMTRCA amended the Atomic Energy Act by directing EPA
to set generally applicable health and environmental standards to govern the stabilization,
restoration, disposal, and control of effluents and emissions at both active and inactive mill
tailings sites.
Title I of UMTRCA covers inactive uranium milling sites, depository sites, and "vicinity
properties," that became contaminated with uranium mill tailings from the uranium milling sites.
In addition to giving EPA responsibilities for setting standards, Title I designated the U.S.
Department of Energy (DOE) as the agency responsible for implementing EPA's standards for
the tailings piles (residual radioactive material) and vicinity properties and for providing long-
term stewardship of the disposal sites. In addition, the U.S. Nuclear Regulatory Commission
(NRC) was designated to review completed site cleanups for compliance with EPA standards
and to license the state or DOE for long-term stewardship of the disposal sites..
Title II of the Act covers operating uranium processing sites licensed by NRC. EPA was directed
to promulgate standards for the processing, possession, transfer, and disposal of uranium mill
tailings (byproduct material). NRC or its Agreement States were required to implement and
enforce these standards at Title II sites.
Thus, 40 CFR Part 192 establishes standards for active and closed mill sites, including ground
water, soil, and building cleanup requirements. These standards are applicable to uranium and
thorium extraction facility licensing, operations, sites, and wastes and are implemented and
enforced by NRC and its Agreement States and DOE. Part 192 applies to residual radioactive
material (Title I only) and byproduct material (Title II)_from conventional mills, ISR facilities,
and heap leach facilities, but not to conventional mines (open pit or underground). Uranium
byproduct material is defined as([§192.31(b)]:
... the tailings or wastes produced by the extraction or concentration of uranium
from any ore processed primarily for its source material content. Ore bodies
depleted by uranium solution extraction operations and which remain
underground do not constitute "byproduct material" for the purpose of this
subpart.
Draft Technical Report 4 Revised Draft-September 2014
-------
Since 40 CFR Parti92 was promulgated, the emphasis in uranium recovery methods has shifted
from conventional milling to ISR, which is considered to be "underground milling."2 Figure 1-1
quantifies this shift in uranium production from conventional milling to ISR (NRDC 2012).
50
45
U.S. Annual Uranium Prouction (million Ibs. U308)
•By-product
•Conventional
• ISL
Source: NRDC 2012
Figure 1-1.
Historical U.S. Uranium Production after World War II
In the ISR process, chemical solutions (i.e., lixiviants) are pumped underground through an array
of wells into the ore body, where the uranium is dissolved. The lixiviants and teachable
constituents are then pumped to the surface, where the uranium is extracted (see Figure 1-2).
Based on the definition presented above, any leaching solutions returned to the ground after
uranium recovery would be byproduct material.
2 Like conventional mills, ISR operations are regulated by the NRC as a form of uranium processing.
However, the injection-extraction technology is also used for the recovery of other minerals, where it is broadly
known as "solution mining." Where this report uses the term "mining," which may be more familiar to the general
public, it is referring to the ISR extraction method. The NRC is constrained by the Atomic Energy Act, as Amended,
from licensing mines. Since it does regulate milling, and the underground chemical processes used to extract
uranium in the ISR process are similar to those for conventional mills, the NRC accordingly licenses those facilities.
Draft Technical Report
Revised Draft - September 2014
-------
Resin Transfer
(Cleaned resin stripped
of uranium)
Uranium in Solution
Yallowcake
Thickener
Spent
Eluate sent
to Disposal
Well or
Evaporation
Pond
Washing and
Dewatering
i
1
Source: http://www.tceq.texas.gov/assets/public/permitting/rad/insitu/in situ diagram.jpg
Figure 1-2. In-situ Uranium Recovery - Process Flow Diagram
In response to this shift in production technology, EPA announced on May 27, 2010, that it
planned to review 40 CFR Part 192. Ground water monitoring within and in the vicinity of an
ISR site serves vital functions that are necessary for efficient uranium recovery with minimal
adverse environmental impacts. Proper monitor well placement and data collection from these
wells ensure that the aquifer constituents are detected and then restored to pre-mining levels.
Without adequate monitor well placement and data collection, including consideration of sample
frequency and sampling time frame, mine operators and regulators (1) may not detect excursions
of lixiviant outside the mining area during operations, and (2) may not be able to confidently
determine whether the affected aquifer needs further restoration or has been restored to its pre-
mining state or another suitable condition that satisfies regulatory requirements.
EPA's standards in 40 CFR Part 192 are required by statute to address nonradiological, as well
as radiological, constituents and to provide for the "protection of human health and the
environment consistent with the standards required under Subtitle C of the Solid Waste Disposal
Act" [UMTRCA sec. 206(b)(2)]. In particular, for Title I sites, UMTRCA states that the
standards shall "... to the maximum extent practicable, be consistent with the requirements of the
Solid Waste Disposal Act, as amended," now known as the Resource Conservation and
Recovery Act (RCRA). For Title II and future NRC-licensed sites, the standards shall be
Draft Technical Report
Revised Draft - September 2014
-------
"... consistent with the standards required under subtitle C of the Solid Waste Disposal Act, as
amended, which are applicable to such hazards."
The existing standards incorporate ground water protection requirements applicable to hazardous
waste management units. These requirements, which are specified in 40 CFR Part 264,
Subpart F, provide a reasonable basis for addressing post-operational ground water monitoring
and restoration at ISR facilities, while also providing the flexibility for site-specific,
performance-based implementation by the regulatory authority (NRC or Agreement State).
Since EPA has not updated its UMTRCA standards since 1983 to cover the ISR technologies, the
NRC had provided regulatory oversight of these facilities through very minor changes to its
10 CFR Part 40 regulations and a series of regulatory guidances (NUREGs) such as NUREG-
1569. In doing so, the NRC and its Agreement States tried to adapt EPA regulations for above-
ground milling to underground operations, including procedures for establishing baseline
conditions, compliance monitoring, determination of Alternate Concentration Limits (ACLs),
site restoration, and corrective actions.
In September 2011, the NRC issued a revision to 10 CFR Part 40 regarding when operators
could commence construction operations, including establishing production, injection, or
monitoring well networks associated with in-situ recovery (Federal Register 2011). Under the
revised regulations, these construction operations cannot begin until a license for handling source
and by-product material is granted. Per 10 CFR 51.4, construction does not include "site
exploration, including necessary borings to determine foundation conditions or other pre-
construction monitoring to establish background information related to the suitability of the site,
the environmental impacts of construction or operation, or the protection of environmental
values." Thus an operator can accumulate background data from exploratory wells, but cannot
develop the detailed data required to establish the baseline conditions within in an ore body prior
to receiving a license from the NRC or an Agreement State.
1.1 Background versus Baseline Wells
Wells are drilled for a variety of purposes during the life cycle of an ISR facility. This section
discusses the terminology used to describe wells used for pre-operational data collection. A
Glossary defining various types of wells and related terms is included as Attachment H.
A key terminology question involves what constitutes "background" and "baseline" wells. In this
document, we retain the term "baseline" and note that it is synonymous with "pre-operational
wellfield background". The term "background" has a precedent from the RCRA arena, and since
UMTRCA calls for consistency with RCRA requirements, there is a strong sense that
"background" should be used for the sake of consistency. On the other hand, "baseline" is the
term used by the NRC, the States and the industry to refer to the pre-operational ground water
chemistry in the wellfield for an ISR operation, and serves as the measure for judging the
adequacy of the post-operation restoration and the regulatory decision to terminate the license. It
is well understood, established in practice for 10-20 years, and used by those involved in ISR
operations and regulation. Use of the term "background" may avoid some confusion in the
RCRA arena, but creates some confusion for those involved in ISR operations and regulations.
Draft Technical Report 7 Revised Draft - September 2014
-------
There are, in fact, many "backgrounds" involved in ISR based on geographical location and time
phases for an ISR operation. The configuration of an ISR operation consists of the wellfield (in
which the ISR operation takes place), surrounded by a ring of monitoring wells, all contained
within a larger area designated as the exempted aquifer, which is, in turn, contained in a larger
aquifer that may be a drinking water aquifer outside the boundaries of the exempted zone. In the
up and down gradient directions outside of the exempted aquifer, there are two "backgrounds"
expected to differ in their respective chemistries. Aquifers above and below the ISR aquifer also
have "backgrounds" that are monitored to detect instances of contamination arising from the ISR
process. How each of these "background" locations plays into the operation and regulation of an
ISR operation is explained below. (The terms up gradient and down gradient are used to
designate locations up and down gradient outside the wellfield where the ISR operation is done).
The characterization and function of various background wells are:
• Non-exempt aquifer up gradient background - The water here should be chemically
oxidizing, with compositions not strongly influenced by the mineralization of the ore
body. These wells should be monitored to detect variations in ground water
compositions throughout the course of the ISR operation to identify and quantify
seasonality effects if present (probably not present in deep aquifers, but may well be
present in near-surface aquifers).
• Non-exempt overlying and underlying aquifer background - These background
monitoring wells would be located in any aquifers immediately above and below the
mined aquifer (i.e., the wellfield) as applicable and monitored, before and during
operations, to detect the occurrence of leaks from the mined aquifer (from pumping
effects or well leaks during operations).
• Exempt aquifer background - This would most probably involve the monitoring well ring
surrounding the wellfield both up and down gradient (and perhaps also a series of down
gradient wells within the exempted aquifer). This "background" would be continuously
monitored prior to and through operations to detect excursions from the wellfield and to
monitor their remediation, and in the down gradient direction to gather the information
necessary to do geochemical modeling of the movement and retardation of contaminants
leaving the wellfield. The down gradient exempt aquifer ground water chemistry should
be different than the up gradient wells, since the up gradient wells should reflect various
degrees of an oxidizing chemical system, whereas the down gradient wells should exhibit
various degrees of a chemically reducing environment.
• Wellfield Background (Baseline) - This is the most important "background"
measurement for an ISR operation. The ore-bearing wellfield is monitored prior to
operations to establish the pre-mining composition of the ground water, and the
monitoring results form the basis for the goals of the restoration phase of the ISR
operation, i.e., returning the system to a state as close as possible to that prior to the
mining. The wellfield wells are also monitored during operations to optimize the
extraction process and potentially detect withdrawal well leaks into the overlying
aquifers. The wellfield pre-mining background is known in the industry as the baseline.
Draft Technical Report 8 Revised Draft - September 2014
-------
• Non-exempt Aquifer Down Gradient Background - This background is measured to
determine the chemical composition of the ground water for comparison with waters
migrating toward it from the restored wellfield and passing through the down gradient
portion of the exempted aquifer. It provides the endpoint for geochemical modeling of the
transport and retardation of contaminants. The geochemical model must demonstrate that
the retardation processes in the down gradient exempted aquifer can reduce contaminants
to the background (or MCL levels) in the down gradient non-exempt aquifer, otherwise
ACLs must be applied.
It is evident that a set of spatially separated "backgrounds" must be measured before operations
to set the requirements for restoration of the wellfield after operations. Monitoring "background"
levels during operations is necessary to detect excursions and leaks into upper and lower
aquifers. Monitoring the wells in the down gradient direction and in the wellfield is necessary to
determine when the wellfield restoration has gone as far as possible to reach pre-operational
conditions.
Throughout this document, we cite many references where the authors use the term baseline to
describe the chemistry of the ground water within the wellfield prior to initiation of leaching
operations. We believe it would be inappropriate and confusing to alter the authors' terminology.
Consequently, in this document, we retain the term "baseline" and note that it is
synonymous with "pre-operational wellfield background."
1.2 ISR Facilities in the United States
As noted above, ISR facilities have become the major source of uranium recovery in the United
States. This section summarizes ISR facilities that have been licensed to operate, that are
currently licensed, or those for which licensing plans are being developed.
NRC states that about 12 ISR facilities exist in the United States
(http://www.nrc.gov/materials/uranium-recovery/extraction-methods/isl-recovery-
facilities.html). Table 1-1 summarizes those ISR sites currently regulated by NRC
(http://www.nrc.gov/info-finder/materials/uranium): the remaining sites are regulated by
Agreement States (mainly Texas). Other Agreement States include Colorado and Utah. Although
Nebraska and New Mexico are also Agreement States, NRC has opted to regulate ISR activities
in those states. Wyoming is not an Agreement State; however, the state imposes regulations such
as ground water monitoring on uranium mines, which may overlap with NRC regulations.
Wyoming also requires operation and closure plans and conducts its own environmental and
safety inspections.
Table 1-1. NRC-Licensed ISR Facilities as of September 2011
Site Name
Crow Butte
Crown Point
Lost Creek
Moore Ranch
Nichols Ranch
Licensee
Crow Butte Resources, Inc.
Hydro Resources, Inc.
Lost Creek ISR, LLC
Uranium One Americas, Inc.
Uranerz Energy Corporation
Location
Chadron, Nebraska
Crown Point, New Mexico
Sweetwater County, Wyoming
Campbell County, Wyoming
Campbell and Johnson Counties, Wyoming
Draft Technical Report 9 Revised Draft - September 2014
-------
Smith Ranch
Willow Creek
Power Resources, Inc.
Uranium One U.S.A.
Douglas, Wyoming (Converse County)
Johnson & Campbell Counties, Wyoming
In addition to the licensed facilities listed in Table 1-1, NRC is considering applications for some
expansions and new facilities as summarized in Table 1-2
(http://www.nrc.gov/materials/uranium-recovery/license-apps.html: uranium-recovery-apps.xls).
Some of these sites have already developed significant background data for their licensing
requests (e.g., Dewey-Burdock in South Dakota, see Attachment A),
Draft Technical Report
10
Revised Draft - September 2014
-------
Table 1-2. Uranium Recovery Facility Applications, Reviews, and Letters of Intent by NRC
ID#
1
2
o
J
4
5
6
7
8
9
10
11
12
13
14
15
20
23
24
25
Company
Uranium One
Cameco (Crow Butte Resources, Inc.)
Cameco (Crow Butte Resources, Inc.)
Lost Creek ISR, LLC
Uranerz Energy Corp.
Uranium One
Uranium One
Powertech Uranium Corporation
Uranium One
Cameco (Crow Butte Resources, Inc.)
Uranium One
Lost Creek ISR, LLC
Strata Energy, Inc.
UR-Energy Corp.
Cameco (Power Resources, Inc.)
Wildhorse Energy
AUC LLC
Cameco (Crow Butte Resources, Inc.)
Cameco (Power Resources, Inc.)
Site
Willow Creek
Crow Butte - North Trend
Crow Butte - Plant Upgrade
Lost Creek
Nichols Ranch
Moore Ranch
Jab and Antelope
Dewey-Burdock
Ludeman
Three Crow
Allemand-Ross
Lost Creek
Ross
Lost Soldier - Amendment
Smith Ranch/Highland CPP
West Alkali Creek
Reno Creek
Marsland
Ruby Ranch
State
WY
NE
NE
WY
WY
WY
WY
SD
WY
NE
WY
WY
WY
WY
WY
WY
WY
NE
WY
Location
Johnson and Campbell Counties
Crawford
Crawford
Sweetwater County
Johnson and Campbell Counties
Converse County
Sweetwater County
Custer and Fall River Counties
Converse County
Dawes County
Converse County
Sweetwater County
Crook County
Sweetwater County
Converse County
Fremont County
Campbell County
Dawes County
Campbell County
Design Type
ISR - Restart
ISR - Expansion
ISR - Expansion
ISR - New
ISR - New
ISR - New
ISR - New
ISR - New
ISR - Expansion
ISR - Expansion
ISR - Expansion
ISR - Expansion
ISR -New
ISR - Expansion
ISR - Expansion
ISR -New
ISR -New
ISR - Expansion
ISR - Expansion
Application Date
Apr-07
Jun-07
Oct-06
Mar-08
Dec-07
Oct-07
Sep-08
Aug-09
Jan-10
Jul-10
Jan-12
Sep-11
Dec-10
Mar-12
FY2011
TBD
Jan-12
Oct-11
FY2013
Status
Code*
5
4
5
5
5
5
o
J
4
3
3
4
Letter of Intent
None
None
None
05/23/07
06/27/07
05/31/07
05/31/07
01/26/07
02/26/09
01/11/10
10/08/10
01/06/10
01/08/10
11/01/10
01/14/10
01/07/10
11/03/10
01/09/10
01/14/10
Status Code: 1 - not received; 2 - acceptance review ongoing; 3 - not accepted, withdrawn, or review postponed; 4 - technical review ongoing; 5 - licensing action completed.
Draft Technical Report
11
Revised Draft - November 26, 2012
-------
The Texas Commission on Environmental Quality (TCEQ) list of Texas ISR sites is presented in
Table 1-3 (http://www.uraniuminfo.org/tceq-list-active-and-pending-permits). This list shows
that there were four active sites, two proposed sites, two sites undergoing closure, and
30 shutdown sites in Texas as of October 2011.3
Additional background on ISR performance is included in Groundwater Modeling Studies at In
Situ Leaching Facilities and Evaluation of Doses and Risks to Off-Site Receptors from
Contaminated Groundwater, Revision 1 (SC&A 2011). That report, revised in May 2012,
evaluates the risk to down gradient receptors who obtain their water from a contaminated well.
Risks are assessed for both radioactive and hazardous chemicals.
3 List provided by Maryann Ryan, Radioactive Materials Division, Texas Commission on Environmental
Quality, October 19, 2011.
Draft Technical Report 12 Revised Draft - November 26, 2012
-------
Table 1-3. List of In-Situ Uranium Mines in Texas
Mine
1. Moser
2. Burns
3. O'Hearn
4. Bruni
5. Lamprecht
6. Pawnee
7. LaPalangana
8. Zamzow
9. Clay West
10. Piedre Lumbre 202
ll.Breluml99
12. Piedre Lumbre 200-201
13. Brelum 106-200
14. Piedre Lumbre 201-205
15. Boots-Brown
16. ElMesquite
17. Holiday
18. Fall City
19. Nell
20. Hobson
21. Longoria
22. Benham
23. Benavides
24. Pawlik
25. Mt. Lucas
26. Trevon
27. McBryde
28. Las Palmas
29. West Cole
30. Tex 1
31. Kingsville Dome
32. Rosita
33. Gray
34. Vasquez
35. AltaMesa
36. Silver Lake
37. La Palangana
38. Goliad Project
Active mines are underlined.
Company
US Steel (USX)
USX
COGEMA
Westinghouse
Intercontinental Energy
Intercontinental Energy
Chevron
Intercontinental Energy
USX
Newfuels
Newfuels
Newfuels
Newfuels
Newfuels
USX
COGEMA
COGEMA
Solution Engineering
Newfuels
Everest
Uranium Resources (URI)
Westinghouse
URI
USX
Everest
Conoco
Caithness
Everest
COGEMA
Everest
URI
URI
Everest
URI
Mestena
Caithness
South Texas Mining Ven.
Uranium Energy Corp
Permit No.
UR01890
UR01890
UR01941
UR01942
UR01949
UR02050
UR02051
UR02108
UR02130
UR02147
UR02148
UR02149
UR02151
UR02152
UR02154
UR02155
UR02156
UR02157
UR02202
UR02208
UR02222
UR02307
UR02312
UR02368
UR02381
UR02407
UR02420
UR02441
UR02463
UR02493
UR02827
UR02880
UR02914
UR03050
UR03060
UR02559
UR03070
UR03075
County
Live Oak
Live Oak
Webb
Webb
Live Oak
Bee
Duval
Live Oak
Live Oak
Duval
Duval
Duval
Duval
Duval
Live Oak
Duval
Duval
Karnes
Live Oak
Karnes
Duval
Bee
Duval
Live Oak
Live Oak
Duval
Jim Hogg
Duval
Webb
Karnes
Kleberg
Duval
Jim Hogg
Duval
Brooks
Jim Hogg
Duval
Goliad
Producing Formation
Oakville
Oakville
Catahoula
Catahoula
Oakville
Oakville
Goliad
Oakville
Oakville
Catahoula
Catahoula
Catahoula
Catahoula
Catahoula
Oakville
Catahoula
Catahoula
No mining
Catahoula
Jackson
Catahoula
Oakville
Catahoula
Oakville
Goliad
Oakville
Oakville
Oakville
Catahoula
Jackson
Goliad
Goliad
Oakville
Oakville
Goliad
Oakville
Goliad
Goliad
Proposed mines are in bold.
Mines undergoing closure are in italics.
All other mines are closed.
Draft Technical Report
13
Revised Draft - November 26, 2012
-------
2.0 RESOURCE CONSERVATION AND RECOVERY ACT
This section provides an overview of the RCRA program. Provisions specifically relevant to ISR
facility licensing and oversight are discussed in detail. These provisions include Subtitle C
facilities, ground water monitoring requirements, and treatment, storage, and disposal facilities
(TSDFs). These descriptions of the RCRA program supply context for the development of
provisions of the Part 192 revision rulemaking. We are required to be "consistent" with RCRA
requirements "to the maximum extent practicable." Some of the specifics in the ISR revision to
Part 192 are derived directly from RCRA program requirements, as noted in later discussions of
these requirements. More specifically, the requirements applied to RCRA subtitle C facilities are
most appropriate to the ISR operations.
2.1 Summary
RCRA was passed in 1976, as an amendment to the Solid Waste Disposal Act of 1965, to ensure
that solid wastes are managed in an environmentally sound manner. RCRA gives EPA the
authority to control hazardous waste from "cradle-to-grave." This includes the generation,
transportation, treatment, storage, and disposal of hazardous waste (Subtitle C). RCRA also
establishes a framework for the management of nonhazardous solid wastes (Subtitle D). Further
amendments to RCRA have extended its application; for example, the 1986 amendments to
RCRA enabled EPA to address environmental problems that could result from underground
tanks storing petroleum and other hazardous substances.
RCRA is a key component of EPA's UMTRCA standards in 40 CFR Part 192. As noted in
Chapter 1, Congress specified that EPA's standards should address nonradiological, as well as
radiological, constituents. Therefore, for Title I sites, UMTRCA states that the standards shall,
"... to the maximum extent practicable, be consistent with the requirements of the Solid Waste
Disposal Act, as amended," now known as RCRA. For Title II and future NRC-licensed sites,
the standards shall be "... consistent with the standards required under subtitle C of the Solid
Waste Disposal Act, as amended, which are applicable to such hazards" [UMTRCA
section 206(a)].
EPA's current standards in 40 CFR Part 192 incorporate the RCRA ground water monitoring
requirements for hazardous waste units specified in 40 CFR Part 264, including statistical
techniques for determining when standards have been achieved. A key question in revising the
current rule is whether, and to what extent, it is appropriate to apply these techniques, which
were developed to address releases to ground water from engineered hazardous waste units, such
as landfills, impoundments, and tanks, to ISR uranium recovery facilities, where the regulated
"unit" is a defined portion of an aquifer under 40 CFR Part 146 (see Section 2.3).
The RCRA approach to protecting ground water represents a reasonable starting point for
developing criteria and standards specific to ISR facilities. The remainder of this chapter
provides additional detail on the RCRA requirements and discusses technical challenges in
applying those requirements to ISR facilities. It should be emphasized that this chapter describes
current RCRA regulations and how they are integrated with 40 CFR 192. It does not explore
changes which EPA may adopt as they revise 40 CFR 192. Such changes are discussed in
Draft Technical Report 14 Revised Draft - November 26, 2012
-------
Section 2.3 below and in various sections of the document that discuss the technical issues
involved in provisions of the ISR rulemaking.
2.2 Ground Water Monitoring Requirements for Treatment, Storage, and Disposal
Facilities
2.2.1 Overview
The ground water monitoring requirements for hazardous waste TSDFs are an important aspect
of the RCRA hazardous waste management strategy for protecting human health and the
environment from accidental releases of hazardous constituents. While land disposal restrictions
and unit-specific standards seek to reduce the toxicity of waste and prevent releases,
respectively, the ground water monitoring requirements represent the last line of defense by
ensuring that any releases are detected and remediated in a timely manner.
TSDFs that manage hazardous waste in landfills, surface impoundments, land treatment units,
and some waste piles (referred to as "regulated units" in RCRA) are required to implement a
ground water monitoring program to detect the release of hazardous constituents to the
underlying ground water. The regulations for permitted facilities are found at 40 CFR Part 264.
Specifically, Subpart F addresses releases from solid waste management units and includes
elements of a monitoring program such as:
• Ground water protection standard
• Hazardous constituents
• Concentration limits
• Point of compliance
• Compliance period
• General monitoring requirements
• Detection monitoring
• Compliance monitoring
• Corrective action
The overall goal of these requirements is to protect the ground water in the uppermost aquifer
(i.e., the aquifer closest to the TSDF) from contamination by the hazardous constituents managed
attheTSDF.
2.2.2 Permitted Facilities
For permitted TSDFs, a ground water monitoring program consists of three phases: detection
monitoring (§264.98), compliance monitoring (§264.99), and corrective action (§264.100). The
phases are sequential, with a facility able to move back and forth between phases as certain
criteria are met. The regulations are written as performance standards that require each facility's
ground water monitoring program to have a sufficient number of wells installed at the
appropriate locations and depths that can yield representative samples of background conditions
and water quality at the point of compliance in the uppermost aquifer (defined as the geological
Draft Technical Report 15 Revised Draft - November 26, 2012
-------
formation nearest the natural surface that is capable of yielding significant quantities of ground
water to wells or springs).
To meet these standards, each facility must design, install, and operate a ground water
monitoring program based on the site's specific geology and hydrology, as well as the type of
waste management unit and the characteristics of the waste being managed. The monitoring
wells must be appropriately designed and installed, and consistent sampling and analytical
procedures must be implemented to ensure accurate and representative samples. The facility's
hazardous waste permit specifies the specific sampling requirements and procedures (including
frequency of sampling).
2.2.3 Detection Monitoring
Detection monitoring is phase one of the ground water monitoring program. In this phase,
facilities are monitored to detect and characterize any releases of hazardous constituents into the
uppermost aquifer. Samples are taken from the monitoring wells and analyzed for specific
indicator parameters and any other waste constituents or reaction products indicating that a
release might have occurred. The facility's permit identifies the specific constituents and
parameters to be monitored and establishes the frequency of sampling. Typically, a sequence of
at least four samples taken at intervals to assure sample independence is required [§264.97(g)].
Samples taken from the point of compliance (i.e., the wells down gradient of the waste
management unit) are compared to the background samples taken from the up gradient well(s).
These samples are analyzed to determine if a statistically significant increase (SSI) in the levels
of any of the monitored constituents has occurred. When analyzing the samples, the facility
owner/operator may use one of the following five methods:
(1) Parametric analysis of variance.
(2) Nonparametric analysis of variance based on ranks.
(3) Tolerance or prediction interval procedure.
(4) A control chart approach.
(5) Another statistical test method approved by the EPA Regional Administrator.
If an SSI is detected, the facility must switch to a compliance monitoring program, unless the
owner/operator can demonstrate that the SSI was due to a sampling analysis, or statistical
analysis error or resulted from natural variations in the ground water chemistry. If unable to
make this demonstration, the owner/operator must:
• Notify the EPA Regional Administrator about the SSI within 7 days.
• Immediately sample all wells for hazardous constituents listed in Part 264, Appendix IX.
• Determine which Part 264, Appendix IX, constituents are present and at what levels.
• Submit an engineering feasibility plan for a corrective action program within 180 days.
• Submit a permit modification application within 90 days to begin a compliance
monitoring program.
Draft Technical Report 16 Revised Draft - November 26, 2012
-------
2.2.4 Compliance Monitoring
The purpose of a compliance monitoring program is to ascertain whether the constituents
released to the uppermost aquifer are exceeding acceptable concentration levels and threatening
human health and the environment. The first step in this process is establishing a ground water
protection standard (GWPS). As stated above, a facility must submit a permit modification
application to switch from detection monitoring to compliance monitoring when an SSI is
detected. As part of this modified permit, the EPA Regional Administrator specifies the GWPS
for the facility. The GWPS establishes:
• The list of hazardous constituents for which to monitor (from Part 261, Appendix VIII).4
• The concentration limits for each of the listed constituents based either on background
levels, Safe Drinking Water Act (SDWA) maximum contaminant levels (MCLs), or
alternate concentration levels determined by the EPA Regional Administrator.
• The point of compliance, which is the vertical surface at which the facility must monitor
the uppermost aquifer to determine if the GWPS is being exceeded.
• The compliance period during which the GWPS applies and compliance monitoring must
be continued.
During compliance monitoring, samples are taken at each well located at the point of compliance
(four samples from each well) and compared to the GWPS. The EPA Regional Administrator
determines the frequency of sampling, which is specified in the modified facility permit. At a
minimum, samples must be taken at least semiannually. The facility must also analyze samples
for 40 CFR Part 264, Appendix IX, constituents at least annually. If any new constituents are
found to have an SSI, then they must be added to the GWPS list of constituents.
If the level of any of the constituents exceeds the GWPS, the owner/operator must notify the
EPA Regional Administrator in writing within 7 days. The owner/operator also must submit a
permit modification application to establish a corrective action program. Compliance monitoring
must continue during this period.
2.2.5 Corrective Action
Once an exceedance of the GWPS has been detected, the facility must act to bring the constituent
concentration levels back into compliance with the GWPS. To achieve this, the owner/operator
must either remove the hazardous constituents or treat them in place. The EPA Regional
Administrator will approve the facility's selected corrective action method and specify the time
frame in which it must take place. Any hazardous constituents that have migrated beyond the
point of compliance also must be remediated. The facility must continue corrective action until
the GWPS has not been exceeded for 3 consecutive years. At that point, the facility may return to
compliance monitoring.
4 A detailed discussion of hazardous constituents that require monitoring is included in Section 3.5,
"Selection of Parameters to Be Used in Groundwater Sampling Programs."
Draft Technical Report 17 Revised Draft - November 26, 2012
-------
2.3 Application to ISR Facilities
While the application of the RCRA ground water monitoring requirements to conventional mills
and tailings impoundments is relatively straightforward, the ISR technology presents additional
technical challenges for post-operational monitoring. First, the technology is applied within the
aquifer by intentionally altering its chemical characteristics to facilitate transport of uranium.
Thus, in the RCRA framework, contaminants have already been released into the environment
and are no longer contained within the engineered hazardous waste unit (e.g., a surface
impoundment).
One perspective on the transference of RCRA approaches to ISR operations is to consider the
mined aquifer production wellfield as the "operational unit," with post-mining restoration
activities considered as "engineering" the unit to prevent migration of contaminants beyond its
boundaries. Post-restoration monitoring of a "restored" aquifer is then the equivalent of post-
closure monitoring around an engineered RCRA disposal facility to assure that contaminants do
not escape from the unit and enter the surrounding environment.
The intent of the operator to release the site for unrestricted use presents the more significant
challenge. Unlike conventional tailings impoundments, which are subject to long-term
stewardship requirements, ISR facilities will leave no significant surface facilities or waste
behind. The ground water will therefore need to be restored throughout the wellfield, which may
show significant heterogeneity. Furthermore, from a corrective action standpoint, the "source" of
contamination cannot necessarily be identified as a specific location within the affected area (the
ore zone). It is therefore particularly important that an appropriate monitoring program be
developed, including an adequate number of wells in the right locations, to determine, with
sufficient confidence, that restoration and long-term stability have been achieved. As discussed
in this document, there may be technical approaches that can be used to modify or extend the
RCRA requirements. Additionally, there may be technical approaches better suited for these
particular types of facilities.
In a further complication for operating ISR facilities, permits for lixiviant injection wells must be
obtained from EPA's Underground Injection Control (UIC) Program developed pursuant to the
SDWA (in some cases, authority to issue UIC permits has been delegated to states). To obtain
the required permit, an operating company submits an application to EPA or a Delegated State
requesting that an aquifer or portion of an aquifer be exempted from protections of the SDWA.
In issuing the UIC permit, the regulatory authority makes a determination whether to grant the
exemption and the extent of exemption in the aquifer affected by the activity (40 CFR Part 146).
The regulatory authority (EPA 40 CFR Part 144) also permits the drilling of the injection wells.
However, it is the NRC or its Agreement States which permit the drilling of production wells,
conversion of injection wells to production wells, and also granting the license for the overall
uranium extraction project to proceed. The primary concern of the UIC regulatory program is
that contaminants not be transported beyond the exempted portion of the aquifer ("excursion")
into an underground source of drinking water (USDW). Requirements for restoration of the
exempted portion of the aquifer under the UIC Program are limited compared to the
requirements of 40 CFR Part 192. Failure to recognize the applicability of 40 CFR Part 192 to all
ground water at an ISR facility (i.e., in the wellfield) has led to a situation in which operators at
Draft Technical Report 18 Revised Draft - November 26, 2012
-------
some ISR facilities have not been held to the more stringent standards in 40 CFR Part 192 (see
case studies included in this document). Furthermore, in some cases, the appropriate baseline
conditions may not have been recorded.
The sections which follow describe the current state of the ISR process and generally how it has
been overseen by the NRC or its Agreement States. These regulatory authorities were given the
responsibility for uranium extraction oversight under UMTRCA and, in the absence of specific
EPA regulatory requirements for ISR facilities, have developed their own procedures and
terminology to comply with the overarching EPA regulations for uranium milling in 40 CFR
Part 192. In some cases, such as for establishing ACLs, the EPA and NRC regulations were not
strictly followed, and no ACL has been established by the NRC to date, even though that agency
has allowed restorations above background or MCLs based on considerations such as class of use
of the water (e.g., is the restored water acceptable for the same uses as the pre-mining water?).
Draft Technical Report 19 Revised Draft - November 26, 2012
-------
3.0 GROUND WATER MONITORING AT ISR FACILITIES
3.1 Overview
The life cycle of an ISR facility typically includes the following ground water-related activities:
• Exploration and development to establish that a commercially viable operation is
possible.
• Establishment of site baseline conditions for ISR (mining) of the ore body.
• Recovery of uranium from the ore body.
• Restoration of the ground water to predetermined conditions.
• Demonstration that restored ground water has reached steady state.
• Post-restoration stability monitoring of the ground water.
• Decommissioning of mined area and surface facilities.
This report is primarily concerned with the pre- and post-operational aspects of ground water
monitoring, specifically establishment of the ground water baseline, demonstration that the
restored ground water has reached steady state, and confirmation through post-restoration
stability monitoring that the ground water quality is not deteriorating over time after restoration.
Figure 3-1 is a graphic representing the evolution of a ground water component of interest during
the phases described below. EPA's 40 CFR Part 192 requires ground water restoration to
background (baseline) or to maximum concentration limits (whichever is higher), and in some
cases allows the regulator to establish an ACL after meeting 19 rigorous listed criteria. However,
NRC has been utilizing a somewhat different standard, termed a "Restoration Goal," for
restoring hazardous constituents in ground water; this standard has not necessarily been
compliant with the EPA regulatory standards. Figure 3-1 shows that the measured post-
restoration ground water concentration is below the Restoration Goal. In practice, this targeted
result may not be realized. This report documents numerous examples where the wellfield was
not returned to baseline conditions. In those examples, the regulatory authorities may have
determined that the deviations from baseline did not impose serious threats to ground water use
ouside the mined area. However, the intention of the 40 CFR 192 rulemaking is to impose a
more systematic and consistent regulatory process compatible with other regulatory regimes
aimed at protecting ground water.
The five phases of ground water monitoring during the life of the ISR facility are:
• Phase 1 - Measure baseline ground water concentrations and establish regulatory
approved restoration goals based on statistical procedures that embrace pre-mining
temporal and spatial variability.
• Phase 2 - Conduct in-situ mining. Detect lixiviant excursions outside the mining area if
they occur. Determine the ground water chemistry at the end of ISR operations.
• Phase 3 - Conduct wellfield restoration. Monitor the progress of restoration through
ground water sampling.
Draft Technical Report 20 Revised Draft - November 26, 2012
-------
Phase 4 - Establish compliance with baseline. During this phase, sufficient wells are
sampled and sufficient samples are collected from each well to statistically compare the
baseline and post-restoration ground water chemistry based on established data quality
objectives. If compliance with the baseline cannot be demonstrated, alternate restoration
goals may be explored with the regulator.
Phase 5 - Conduct long-term stability monitoring. During this phase, use statistical tests
to show that concentration of the monitored species is not increasing with time and that
concentration is not statistically different from baseline conditions, or if baseline
conditions are unachievable, that the concentration is not statistically different from
approved restoration goals.
Measured
Ground
Water
Concentration
1.2 -r
1 "
0.8 "
0.6 --
0.4 --
0.2-1
Date
Phase 1 - Measure baseline ground water concentrations
and establish regulatory restoration values.
Phase 2 - Conduct in-situ mining.
Phase 3 - Conduct wellfield restoration.
Phase 4 - Establish compliance with baseline.
Phase 5 - Conduct long-term stability monitoring.
Figure 3-1. Variation of Typical Ground Water Constituent over Time
3.2 Pre-operational Monitoring (Phase 1)
The key to any baseline monitoring program is to adequately characterize temporal and spatial
variations in ground water within the ore zone before mining begins. In order to provide the basis
of comparison for assessing progress in restoring the wellfield after mining has been completed,
the breadth of pre-operational ground water monitoring needs to be sufficiently robust for
adequate statistical comparisons with post-operational monitoring.
Draft Technical Report
21
Revised Draft - November 26, 2012
-------
3.3 The ISR Leaching Process (Phase 2)
During typical ISR operations, chemicals such as sodium carbonate/bicarbonate and gaseous
oxygen/hydrogen peroxide are added to the ground water to produce a concentrated oxygen-rich
leaching solution called the lixiviant. The lixiviant is injected into the production zone to create
ground water oxidizing conditions, which mobilize the uranium from the uranium-rich geologic
zone. This mobilized uranium is pumped back to the surface for extraction at a processing plant
(Figure 3-2).
Injection Well— *-
/
/
^^P
\ ;
\
i
_j
/
i
\
f
\
I
s:
\
'/
i
1 Potentiometric Surface (Exaggerated)
. Less Permeable Strata " ~^'
__^_ Ore Bearing Sand
^•""- •*- ^ "^Z>*.
-" , : ~^ ^
^-'"^ "--- -~r»
^.^^ -—--"^
z.~~* »
• Less Permeable Strata .
*
t
:?
— — -
— *- Production Well
^
^!r^--^--------^j--
-^ Perforations
Figure 3-2. Idealized Schematic Cross Section to Illustrate Ore-Zone Geology and
Lixiviant Migration from an Injection Well to a Production Well (NRC 2009)
The most common injection/pumping patterns are five- and seven-spot (NRC 2003). The shape
of the mineralized ore body and surface topography, however, may give rise to other patterns
(NRC 1997). A typical five-spot pattern contains four injection wells and one centrally located
recovery well. The dimensions of the pattern vary depending on the mineralized zone, but the
injection wells are generally between 40 to 150 feet apart. To effectively recover the uranium
and also to complete the ground water restoration, the wells are often completed so that they can
be used as either injection or recovery wells. During mining operations, a slightly greater volume
of water will be recovered from the mineralized zone aquifer than was injected, in order to create
a cone of depression or a flow gradient towards the recovery wells. This practice is intended to
minimize excursions of leachate outside the production area. Ground water monitoring is
necessary to detect any excursions of lixiviant outside the mining area during operations. Figure
3-3 shows typical well arrangements using five- and seven-spot patterns. Figure 3-4 illustrates a
Draft Technical Report
22
Revised Draft - November 26, 2012
-------
typical wellfield. Piping connecting the individual wells to the header house is often run
underground.
Ore body size and geometry will also influence the number of wells in a wellfield. For
example, at the Crow Butte ISL facilities in Dawes County, Nebraska, the number of
injection and production wells varied from about 190 in the first wellfield (MU-1) to
about 900 in later wellfields (MU-5 and MU-6) (NRC 1998).
Four types of wells predominate at uranium ISR facilities during the operational (leaching) phase
(see Figure 3-3):
(1) Injection wells for introducing solutions into the uranium mineralization.
(2) Production wells for extracting uranium-enriched solutions.
(3) Perimeter excursion monitoring wells for assessing containment of leachate within the
wellfield (the ore zone monitor wells in Figure 3-3).
(4) Excursion monitoring wells in the overlying (and/or underlying) aquifer
A
Recovery Trunkline
/ Welffietd Building
A
5-Spot Pattern
O
Injection Weil Recovery Well
(Located at Each {Located at Each
Grid Intersection) Grid Center)
Patterns Repeal Through Wellfield
O
O
A
A
• Injector Recovery Weils
& Ore Zone Monitor Wells
O Shallow Zone Monitor Weils
(One Per 4 Acres)
A
Inset: 5-Spot Pattern
y_
I X
Groundwater Flow
Figure 3-3. Schematic Diagram of a Wellfield Showing Typical Injection/Production
Well Patterns, Monitoring Wells, Manifold Buildings, and Pipelines (NRC 2009)
Draft Technical Report
23
Revised Draft - November 26, 2012
-------
Some of these wells will be used to define initial baseline conditions, to monitor the progress of
restoration, and to determine whether long-term stability has been achieved.
Injection wells at ISR facilities are defined as Class III wells and are regulated under
40 CFRPart 146, "Underground Injection Control Program: Criteria and Standards." This
regulation establishes construction, operating, and monitoring requirements that must be
approved by EPA. EPA only has permitting authority over the injection wells, whereas NRC or
its Agreement States authorize the drilling of production wells and conversion of injection wells
to production or monitoring wells.
Figure 3-4. Wellheads and Header House, Smith Ranch, Converse County, Wyoming
(NRC 2009, Figure 2.1-4)
3.3.1 Excursions during Operations
As noted elsewhere, the focus of this report is on ground water monitoring to establish that an
ISR operation is in compliance with regulatory requirements. Most of the monitoring efforts
ultimately contribute to a determination that the ground water conditions in the mined unit have
been restored to pre-mining levels or to acceptable levels consistent with license conditions. In
addition, licensing conditions typically require monitoring in wells around the periphery of the
mined unit, as well as monitoring wells in overlying aquifers, to detect excursions of production
fluids from the ISR operation into the surrounding ground waters in the mined aquifer and
surrounding aquifers. This section briefly discusses monitoring during operational phases to
detect excursions. Figure 3-3 shows an ore zone surrounded by a ring of horizontal perimeter
monitor wells (the triangular symbols). The figure also shows the location of monitor wells in an
overlying aquifer (open circles). Generally, the density of these vertical sampling (monitoring the
overlying aquifers) wells is much lower than that of the horizontal perimeter monitoring wells.
The purpose of all these wells is to detect excursions of production fluids from the ore zone
Draft Technical Report
24
Revised Draft - November 26, 2012
-------
during and after leaching operations. The spacing of horizontal excursion monitoring wells
(i.e., wells within the production aquifer) is based on site-specific conditions, but typically they
are about 300-500 feet apart. The distances between monitoring wells and the distances of
monitoring wells from the wellfield are generally similar. The specific location and spacing of
the monitoring wells is established on a site-by-site basis by license condition. The spacing is
often modified according to site-specific hydrogeologic characteristics, such as the extent of the
confining layer, hydraulic gradient, and aquifer transmissivity (NRC 2009).
To fully understand the capabilities of perimeter monitor wells in detecting excursions, their
locations relative to the wellfield should be supported with ground water transport modeling. As
described in the Wellfield Restoration Report Christensen Ranch Project Wyoming
(COGEMA 2008a, Section 8.2.2.1):
Groundwater velocities were calculated for each of the MUs based on hydraulic
conductivity, natural hydraulic gradient and porosity estimates. Travel times were
calculated for the time to reach the monitor ring (typically 400 feet from the
wellfield edge) ... The MURDPs [restoration data packages] present the data used
in the calculations. The range of groundwater velocity determined from those
calculations was from 0.0088 to 0.043 ft/d (3.2 to!5.5ft/yr). Estimated travel
times to reach the 400 foot monitor well ring ranged from 26 to 123 years.
SC&A has made similar ground water flow simulations at many generic sites (SC&A 2011).5
The SC&A calculations show similar travel times under comparable hydraulic conditions to
those presented by COGEMA. Thus, excursions associated with normal advective flow would
not be detected over the lifetime of ISR facilities with comparable hydraulic properties.
NRC requires that three species be specified as excursion indicators (e.g., chloride, conductivity,
and alkalinity) (see also Section 3.5.5) and deems that an excursion occurs when two of the
indicators exceed their established upper control limits (NRC 2003). As described in NRC's
Standard Review Plan (NRC 2003, p. 5-41):
Upper control limits for a specific excursion indicator should be determined on a
statistical basis to account for likely spatial and temporal concentration
variations within the mineralized zone. Statistical techniques, such as the
student's t-test, are acceptable for setting upper control limits. In some cases, the
use of a simple percentage increase above baseline values is acceptable. The staff
has decided that in areas with good water quality (a total dissolved solids less
than 500 mg/L), setting the upper control limit at a value of 5 standard deviations
above the mean of the measured concentrations is an acceptable approach.
However, in some aquifers of good water quality, low chloride concentrations
have been found to have such a narrow statistical distribution that a specified
concentration (e.g., 15 mg/L) above the mean or the mean plus 5 standard
deviations approach, which ever is greater, has been used to establish the
chloride upper control limit.
5 Revision 2 of this report was issued on May 8, 2012.
Draft Technical Report
25
Revised Draft - November 26, 2012
-------
In practice, establishment of upper control limits may differ from the guidance provided in the
Standard Review Plan. For example, in the 2011 draft license renewal at Crow Butte (Crow
Butte 2011), the operator is required to collect four samples from each perimeter monitoring
well, with at least 14 days between sampling, to establish a baseline for that well. The upper
control limits (UCLs) for each of the three indicator parameters are then set equal to the
maximum baseline value plus 20%. For parameters with baseline concentrations that average
50 milligrams per liter (mg/L) or less, the operator can chose the maximum baseline plus 20%,
the baseline average plus 5 standard deviations, or the baseline average plus 15 mg/L. This is
only one approach that has been used at ISR sites. In an earlier operation at Christensen Mine
Unit 6 (COGEMA 1996), the UCLs were based on the average of all perimeter wells plus
5 standard deviations. For the chloride excursion indicator, the NRC license allowed use of either
the mean plus 5 standard deviations or the mean plus 15 mg/L, whichever was higher. This
option was based on the fact that the mean chloride concentration was about 5 mg/L with a
standard deviation of less than 1 mg/L. Consequently, use of too tight a UCL could result in false
positive indications of lixiviant excursions.
Excursions at operational ISR sites are common. Staub et al. (1986) summarized information on
excursions at eight sites (seven in Wyoming and one in Texas) developing during the early years
of in-situ leaching. Because this study was done early in the use of ISR, several of the sites were
experimental rather than full-scale production operations. The authors noted that:
Despite inconsistencies in identifying excursions it is evident that many
excursions did occur. Most horizontal excursions were brought under control
quickly. However, wells used to monitor for vertical excursions were on excursion
status repeatedly and for excessively long periods of time. In many cases
restoration procedures were eventually required. It is particularly important to
recognize vertical excursions at an early stage in order to avoid costly and time
consuming restoration.
The relative intractability of vertical excursions emphasizes the importance of determining that
all abandoned boreholes in the area are investigated to ensure that they are properly sealed and
that adequate testing is done to fully characterize the local hydrogeology, particularly continuity
of the aquitards. Testing the integrity of the production wells against leakage into overlying (or
underlying) aquifers is also a concern, since remediation of excursions into these aquifers can be
difficult. This also speaks to the importance of having a sufficient number of monitoring wells in
the overlying aquifers, so that excursions can be detected early and corrective action taken to
avoid the need for extensive restoration activities.
3.3.1.1 Case Histories
The examples of actual excursions provided here are not intended to be a comprehensive list of
all excursions that have been documented, but rather to provide examples of site-specific
excursions and show the periods over which they persist and how they are addressed.
Draft Technical Report 26 Revised Draft - November 26, 2012
-------
Crow Butte
The Crow Butte compliance log shows that as of June 2008, about 20 excursions had been
reported since mid-1999 (Cohen 2008).
For example, during regularly scheduled biweekly testing of monitor wells at Crow Butte, an
excursion in perimeter monitor well CM8-21 was detected when two of the three required
indicators (alkalinity and chloride) exceeded the respective UCLs (Teahon 2006a). If a single
indicator parameter exceeds the UCL, then a somewhat higher control limit is allowed (the single
indicator control limit or SCL, than if two or more indicators exceed the multiple indicator
control limit.6 As required by the facility's NRC license, Crow Butte staff resampled the well
within 48 hours and found that the UCLs were still exceeded. Again, as required by NRC
license, Crow Butte instituted weekly sampling and unspecified corrective action. Weekly
sampling continued from January 24 through April 4, 2006. Samples taken from February 28
through April 4 were below the UCLs, and the well was removed from excursion status. Figure
3-5, Figure 3-6, and Figure 3-7 present charts showing the behavior of the excursion indicators
during the excursion of Well CM8-21. Horizontal lines in these figures indicate the control
limits: SCL, single indicator control limit, and MCL, multiple indicator control limit. The SCLs
are 20% higher than the MCLs.
Results for a similar excursion in Crow Butte Well CM9-16 are reported in Teahon 2005a.
In another example at Crow Butte, Monitor Well PR-15 was observed to exceed the multiple
parameter UCLs for chloride and conductivity on September 6, 2006. This well was a baseline
restoration well for Mine Unit 1, which was also being used as a perimeter monitoring well for
Mine Unit 2 (Teahon 2006b). The operator noted that restoration activities in Mine Unit 2
adjacent to PR-15 included ground water transfer and wellfield recirculation. Two other baseline
restoration wells, IJ-13 and PR-8, from Mine Unit 1, have remained on excursion status since
December 27, 2002, and December 23, 2003, respectively. Because of the geometry of Mine
Units 2 and 3, the operator believed that PR-15 will continue to exhibit the same trend as IJ-13
and PR-8 until Mine Units 2 and 3 can be fully restored along the perimeter of Mine Unit 1. The
increases in chloride and conductivity were associated with a drop in the water level of
Well PR-15, presumably a result of ground water transfer.
Griffin (2005) discusses problems with shallow monitor Well SM6-28 (i.e., a well in an aquifer
overlying the aquifer containing the ore zone). Increases in conductivity and alkalinity were
detected in this well on June 16, 2005. Crow Butte Resources (CBR) believed that this apparent
excursion was due to increased ground water levels caused by the significant precipitation
received at the facility in the spring of 2005 and was not caused by mining activity. This
conclusion was supported by the fact that the water level in the well increased 4 feet during the
spring and was within 10 feet of the top of the casing at the well. Ground water quality in the
area is under the influence of surface water.
6 Exclusively in this section, MCL is the "multiple parameter control limit" and should not be confused
with the "maximum contaminant level" used elsewhere in this report.
Draft Technical Report 27 Revised Draft - November 26, 2012
-------
As the water level dropped in Well SM6-28, the excursion indicator parameters declined.
Consecutive weekly samples taken on June 21, June 28, and July 5, 2005, showed that the
indicators had quickly recovered to values below the MCLs, and the well was removed from
excursion status (Teahon 2005b).
CM8-21
I
I
Alkalinity
500-
450
400
350-
300
250
200
150-
ur\
f
* * * * \> * * ~« »--•"*""*
• • "%> *o **o ° ° ° ° ° ° ^ Jj° •& -i. \ \
'\ ^ J> ^°
//S///S////S///S////
Cute
- AlkMCL . Alkalinity —Alk SCL
Figure 3-5. Alkalinity (mg/L) Variation during Excursion of Crow Butte
Monitor Well CM8-21
CM8-21
Conductivity
2750
1750-
750 •-
250
\\
H 1 1-
//////S/S//////S///S
Cooa MCL Conductwty Cond SCL
Figure 3-6. Conductivity (jimho/cm) Variation during Excursion of Crow Butte
Monitor Well CM8-21
Draft Technical Report
28
Revised Draft - November 26, 2012
-------
CM8-21
Chloride
300
250
200
100
* %
Cl MCL • CNonde CI SCL
Figure 3-7. Chloride (mg/L) Variation during Excursion of Crow Butte
Monitor Well CM8-21
Christensen Ranch
Excursion monitor wells continue to be sampled after production operations have ceased.
Monitoring may continue through the restoration and stabilization phases. At Christensen Ranch,
sampling of the perimeter monitor wells in Mine Units 2 through 6 was conducted monthly
during restoration and quarterly thereafter. Except for those in Mine Unit 5, these perimeter
monitoring wells were unaffected during restoration and post-restoration (through March 2008).
In Mine Unit 5, one well (5MW66) went into an excursion mode on July 21, 2004, 1 month
before the final planned round of stabilization sampling (COGEMA 2008b). This well is directly
down gradient from one of the production modules within Mine Unit 5. The excursion was
finally terminated on February 23, 2011, using corrective pumping, but monitoring continued
until April 11, 2011 (Arbogast 2011 a).
During the second quarter of 2011, three other wells (2MW89, 4MW1 and 5MW8) were also on
excursion status at Christensen Ranch, but the excursion durations were significantly shorter than
for Well 5MW66 (Arbogast 201 Ib). Table 1-3 summarizes the details of these additional
excursions.
Draft Technical Report
29
Revised Draft - November 26, 2012
-------
Table 3-1. Wells on Excursion at Christensen Ranch - 2Q 2011
Well
2MW89
4MW1
5MW8
Location
Perimeter well in
restored wellfield of
Mine Unit 2
Perimeter well in
restored wellfield of
Mine Unit 4
Perimeter well in
restored wellfield of
Mine Unit 5
On Excursion
March 9, 2011
March 23, 20 11
April 19, 2011
Off Excursion
May 2, 2011
May 16,2011
May 31, 2011
Corrective Action
Pumping two
adjacent recovery
wells
Pumping one
adjacent recovery
well
Pumping one
adjacent recovery
well
It is noteworthy that these excursions were detected by perimeter monitor wells in restored
wellfields for which the operator was seeking restoration approval. Attachment B provides
additional details.
3.3.1.2 Summary of Excursion Experience in ISR Operations
Excursions of production/restoration waters into monitoring wells within the mined aquifer and
overlying aquifers have been observed often and present a problem for the operator and regulator
during the period of regulatory control. In most cases, the excursions were quickly restored in the
mined unit, which is usually an exempted aquifer, and the excursions appear to pose little
long-term potential for contamination to non-exempt ground water down gradient beyond the
limits of the exempted area. Excursions into overlying, non-exempt aquifers pose a more
significant concern and may require lengthier restoration efforts to remediate the problem, which
emphasizes the need for monitoring in the overlying aquifers and measures to test the integrity of
the wells against leakage.
3.4 Post-operational Monitoring (Phases 3 through 5)
The intent of restoration efforts is to establish hydrologic and geochemical conditions in the
mined areas that will maintain steady-state conditions in all potentially affected aquifers
(i.e., overlying, underlying, and adjacent aquifers) and ensure that there is no degradation of
water quality from pre-mining conditions. During restoration, the operator monitors progress by
periodic sampling and analysis of the ground water constituents to determine when steady-state
conditions are attained. Establishing steady-state conditions requires that the ground water
potentiometric surface be restored, to the extent practicable, to its preleaching status, so that the
flow regime is similar to that existing before mining. In addition, constituents in the ground
water must be returned to the predetermined restoration goals and remain at that level for a
sufficient period to demonstrate that the results are not trending upwards to higher concentration
levels.
Once the operator concludes that restoration is complete and has obtained concurrence from the
regulator(s) that a steady state has been established, post-restoration stability monitoring begins.
The purpose of the stability monitoring is to demonstrate that the aquifer conditions established
at the end of restoration are sustainable over time. Currently, the duration of the stability
monitoring period is site specific, with the period established in the license(s). In the past, the
license-established post-restoration period typically was about 6 months. More recently, the
Draft Technical Report
30
Revised Draft - November 26, 2012
-------
trend has been to increase the stability monitoring period established in the license. In practice,
the actual period of stabilization may be several years, based on iterative analyses of additional
samples requested by the regulators.
3.5 Selection of Parameters to Be Used in Ground Water Sampling Programs
This section discusses considerations involved in the selection of parameters to be included in
ground water monitoring programs at ISR facilities. The focus is on parameters required for
baseline and post-restoration compliance monitoring, but parameters required for excursion
monitoring and ground water modeling are also included. A site-specific example is provided
illustrating how the required monitoring list may be winnowed over time based on field
measurements and regulatory approval.
3.5.1 Regulated Constituents
3.5.1.1 EPA Regulations
Various EPA regulations establish parameters in the ground water that may require monitoring.
Current EPA regulations at 40 CFR Part 192 define constituents to be monitored during
processing of uranium ores based on RCRA regulations (see Section 2.0 for a detailed discussion
of the RCRA regulations). The standards are presented in §192.32 and are referenced to relevant
RCRA regulations there. According to §192.32:
(2) Uranium byproduct materials shall be managed so as to conform to the
ground water protection standard in §264.92 of this chapter, except that for the
purposes of this subpart:
(i) To the list of hazardous constituents referenced in §264.93 of this chapter are
added the chemical elements molybdenum and uranium ...
The regulation in §264.93, "Hazardous constituents," states that:
(a) The Regional Administrator will specify in the facility permit the hazardous
constituents to which the ground-water protection standard of §264.92 applies.
Hazardous constituents are constituents identified in appendix VIII of part 261 of
this chapter that have been detected in ground water in the uppermost aquifer
underlying a regulated unit and that are reasonably expected to be in or derived
from waste contained in a regulated unit, unless the Regional Administrator has
excluded them under paragraph (b) of this section.
The RCRA regulations, in turn, provide in Appendix VIII to 40 CFR Part 261 an extensive list of
hazardous constituents for which the EPA Regional Administrator may specify ground water
monitoring. Most of the items listed in Appendix VIII are complex organic chemicals such as
aldrin; chlordane; lindane; 2,4-D; and DDT. Inorganic species are listed in Table 3-2. The
Regional Administrator will include those hazardous constituents from Appendix VIII that have
been detected in the ground water. As noted above, EPA added [via §192.32(a)(2)(i)] uranium
Draft Technical Report 31 Revised Draft - November 26, 2012
-------
and molybdenum7 to this list of hazardous constituents. Table 1 of §264.94 defines specific
concentration limits for the following hazardous constituents: arsenic, barium, cadmium,
chromium, lead, mercury, selenium, silver, and several organic species such as lindane. To the
maximum concentration limits in Table 1 of §264.94, EPA added [via §192.32(a)(2)(ii)]
combined radium (Ra-226) + Ra-228 and gross alpha (excluding radon and uranium).
For those hazardous constituents for which a maximum concentration is not specified, the
concentration in the ground water cannot exceed the background level [§264.94(a)(l)].
Additional hazardous constituents from Appendix VIII relevant to ISR facilities for which the
background level applies include nickel, molybdenum, thallium, uranium, and vanadium
pentoxide (see Table 3-2).
The standards in §192.32(a)(2)(iii) also establish a requirement for a monitoring program as
specified in §264.98. Section 264.98 references Appendix IX to Part 264 as the source list for
hazardous constituents. Appendix IX to Part 264 is similar to Appendix VIII to Part 261 but also
includes cobalt, sulfide, tin, vanadium (rather than V2Os), and zinc.
EPA's National Primary Drinking Water Regulations (NPDWR) specify MCLs for antimony,
arsenic, barium, beryllium, cadmium, chromium, copper, fluoride, lead, mercury, nitrate, nitrite,
selenium, thallium, gross alpha, beta + gamma, Ra-226 + Ra-228, and uranium
(40 CFR Part 141). The MCL is defined as the maximum permissible level of a contaminant in
water that is delivered to any user of a public water system. Public water systems have at least
15 service connections or regularly serve an average of at least 25 individuals daily for at least
60 days out of the year. In addition, 40 CFR Part 141 establishes maximum contaminant level
goals (MCLGs), which are the maximum levels of a contaminant in drinking water at which no
known or anticipated adverse effect on the health of persons would occur, and which allows an
adequate margin of safety. MCLGs are nonenforceable health goals.
A Note on Concentration Levels/Limits
According to the NPDWR at §141.2: Maximum contaminant level
means the maximum permissible level of a contaminant in water which
is delivered to any user of a public water system.
Maximum concentration level or limit is not a defined term under
RCRA. Rather §264.94 - Concentration Limits refers to the maximum
concentration of constituents.
Conditions for establishing alternate concentration limits are delineated
in §264.94(b).
7 EPA recognized, when it was writing the original uranium milling regulations, the fatal impact of
molybdenum on young bovine calves, likely to be present in vicinity of uranium mills or consuming water
contaminated from mill effluent.
Draft Technical Report
32
Revised Draft - November 26, 2012
-------
EPA's National Secondary Drinking Water Regulations (NSDWR or secondary standards) are
nonenforceable guidelines regulating contaminants that may cause cosmetic effects (such as skin
or tooth discoloration) or aesthetic effects (such as taste, odor, or color) in drinking water
(40 CFR Part 143). EPA recommends secondary standards for water systems but does not
require systems to comply. However, states may choose to adopt them as enforceable standards.
Analytes that may be relevant to ISR facilities included under the NSDWR include aluminum,
chloride, copper, fluoride, iron, manganese, pH, silver, sulfate, total dissolved solids (IDS), and
zinc.
EPA also has in place a program to designate additional species that might be included under the
NPDWR. The SDWA directs EPA to publish a list of contaminants (referred to as the
Contaminant Candidate List, or CCL) to assist in priority-setting and to determine whether to
regulate these contaminants with an NPDWR. EPA has determined that the following species
initially on the CCL do not need an NPDWR: sodium, sulfate, manganese, and boron. Species
that are included on the CCL 3 list, for which determinations have not yet been made regarding
NPDWRs, include cobalt, molybdenum, germanium, strontium, tellurium, and vanadium
(http ://water. epa.gov/scitech/drinkingwater/dws/ccl/ccl3. cfm).
Injection wells at ISR facilities are regulated under 40 CFR Part 146. At §146.3, this regulation
defines USDW as an aquifer or its portion:
(l)(i) Which supplies any public water system; or
(ii) Which contains a sufficient quantity of ground water to supply a public
water system;
and
(A) Currently supplies drinking water for human consumption; or
(B) Contains fewer than 10,000 mg/l total dissolved solids; and
(2) Which is not an exempted aquifer.
If the USDW supplies a public water system as described above, then the drinking water
standards at 40 CFR Part 141 are applicable. However, if the aquifer is an exempted aquifer, then
Part 141 does not apply. To qualify as an exempted aquifer, the following criteria must be met
(§146.4):
(a) It does not currently serve as a source of drinking water; and
(b) It cannot now and will not in the future serve as a source of drinking
water because:
(1) It is mineral, hydrocarbon or geothermal energy producing, or can
be demonstrated by a permit applicant as part of a permit
application for a Class II or III operation to contain minerals or
hydrocarbons that considering their quantity and location are
expected to be commercially producible.
(2) It is situated at a depth or location which makes recovery of water
for drinking water purposes economically or technologically
impractical;
Draft Technical Report 3 3 Revised Draft - November 26, 2012
-------
(3) It is so contaminated that it would be economically or
technologically impractical to render that water Jit for human
consumption; or
(4) It is located over a Class III well mining area subject to subsidence
or catastrophic collapse; or
(c) The total dissolved solids content of the ground water is more than 3,000
and less than 10,000 mg/l and it is not reasonably expected to supply a
public water system.
Table 3-2 summarizes the species that may require monitoring under various EPA regulations
described above and indicates limits specified in the regulations. It should be noted that some
differences exist between the RCRA regulatory limits in Table 1 of 40 CFR 264.94 and the
MCLs in the National Drinking Water Regulations at 40 CFR Part 141. The table also lists
nonmandatory MCLGs below which no health effects are expected and the nonmandatory MCLs
for NSDWRs at 40 CFR Part 143.
Draft Technical Report 34 Revised Draft - November 26, 2012
-------
Table 3-2. Ground Water Species Identified in EPA Regulations That May
Require Monitoring at ISR Facilities
Species
EPA Regulation
Regulatory Concentration Limit (mg/L)
40 CFR 264.94
(Table 1)
40 CFR 141
(MCL)b
40 CFR 141
(MCLG)
40 CFR 143
(MCL)C
Molybdenum
Uranium
Ra-226 + Ra-228
Gross alpha
(ex. U and radon)
Nickel
Thallium
Vanadium Pentoxide
Arsenic
Barium
Cadmium
Chromium
Cobalt
Lead
Mercury
Selenium
Silver
Sulfide
Tin
Antimony
Beryllium
Copper
Fluoride
Nitrate
Nitrite
Total nitrate plus nitrite
Beta + gamma
Aluminum
Chloride
Iron
Manganese
pH
Sulfate
Total dissolved solids
Zinc
Vanadium
§192.32(a)(2)(i)
§192.32(a)(2)(i) 740 CFR 141
§192.32(a)(2)(ii) 740 CFR 141
§192.32(a)(2)(ii) 740 CFR 141
§264.93/§261, Appendix VIII
§264.93/§261, Appendix VIII/
40 CFR 141
§264.93/§261, Appendix VIII
§264.94, Table 1/40 CFR 141
§264.94, Table 1/40 CFR 141
§264.94, Table 1/40 CFR 141
§264.94, Table 1/40 CFR 141
§264.98 (Appendix IX)
§264.94, Table 1/40 CFR 141
§264.94, Table 1/40 CFR 141
§264.94, Table 1/40 CFR 141
§264.94, Table 1/40 CFR 141
§264.98 (Appendix IX)
§264.98 (Appendix IX)
40 CFR 141
40 CFR 141
40 CFR 141/40 CFR 143
40 CFR 141/40 CFR 143
40 CFR 141
40 CFR 141
40 CFR 141
40 CFR 141
40 CFR 143/CCL
40 CFR 143
40 CFR 143
40 CFR 143
40 CFR 143
40 CFR 143
40 CFR 143
40 CFR 143/§264.98
(Appendix IX)
CCL/§264.98 (Appendix IX)
0.03
5 (pCi/L) 5 (pCi/L)
15 (pCi/L)
0.002
0.05 0.010
1.0 2
0.01 0.005
0.05 0.1
0.05 0.015
0.002 0.002
0.01 0.05
0.05
0.006
1.3
4
10
(as nitrogen)
1
(as nitrogen)
10
(as nitrogen)
< 4 mrem/yr
dose
4 (MRDL)a
0
0
0.0005
0
2
0.005
0.1
0
0.002
0.1
0.006
0.004
1.3 1.0
4 2.0
10
(as nitrogen)
1
(as nitrogen)
10
(as nitrogen)
0.05 to 0.2
250
0.3
0.05
6.5-8.5
250
500
5
a - MRDL (maximum residual disinfectant level)
b - mandatory MCL per NPDWR
c - non-mandatory MCL per NSDWR
Draft Technical Report
35
Revised Draft - November 26, 2012
-------
3.5.1.2 NRC Requirements
According to NRC's "Standard Review Plan for In Situ Leach Uranium Extraction License
Applications," Section 2.7.3(4), which describes the acceptance criteria for site characterization
hydrology (NRC 2003):
Reasonably comprehensive chemical and radiochemical analyses of water
samples, obtained within and at locations away from the mineralized zone (s),
have been made to determine pre-operational baseline conditions. Baseline water
quality should be determined for the mineralized and surrounding aquifers. These
data should include water quality parameters that are expected to increase in
concentration as a result of in situ leach activities and that are of concern to the
water use of the aquifer (i.e., drinking water, etc.). The applicant should show
that water samples were collected by acceptable sampling procedures, such as
American Society for Testing and Materials D4448 (American Society for Testing
and Materials, 1992).
For example, in situ leach operations are not expected to mobilize aluminum, and
unless an ammonia-based lixiviant is used, ammonia concentrations in the ground
water should not increase as a result of in situ leach operations. Therefore, little
is gained by sampling these parameters. Studies have shown that thorium-230 is
mobilized by bicarbonate-laden leaching solutions. However, studies have also
shown that after restoration, thorium in the ground water will not remain in
solution because the chemistry of thorium causes it to precipitate and chemically
react with the rock matrix (Hem, 1970). As a result of its low solubility in natural
waters, thorium is found in only trace concentrations. Additionally, chemical tests
for thorium are expensive, and are not commonly included in water analyses at in
situ leach facilities.
Section 5.7.8.3 of NRC 2003 states:
The applicant should identify the list of constituents sampled for baseline
concentrations. Table 2.7.3-1 [see Column 3, Table 3-3 of this report] provides a
list of acceptable constituents for monitoring at in situ leach facilities.
Alternatively, applicants may propose a list of constituents that is tailored to a
particular location. In such cases, sufficient technical bases must be provided to
demonstrate the acceptability of the selected constituent list.
3.5.1.3 State of Texas Requirements
o
TCEQ provides guidance on ground water analyses. This guidance specifies measurement of the
26 parameters listed in the following table:
Texas Administrative Code, Title 30, Part 1, Chapter 331, SubchapterF, Rule §331.104.
Draft Technical Report 36 Revised Draft - November 26, 2012
-------
Figure: 30 TAC §331. 104(b)
Calcium (Ca) in mg/L
Magnesium (Mg) in mg/L
Sodium (Na) in mg/L
Potassium (K) in mg/L
Carbonate (CO3) in mg/L
Bicarbonate (HCO3) in mg/L
Sulfate (SO4) in mg/L
Chloride (Cl) in mg/L
Nitrate [NO3, as nitrogen (N)] in mg/L
Fluoride (F) in mg/L
Silica (SiO2) in mg/L
Total Dissolved Solids (TDS) in mg/L
Electrical Conductivity (EC) in mhos/cm
Alkalinity (Alk) in standard
units
pH in standard units
Arsenic (As) in mg/L
Cadmium (Cd) in mg/L
Iron (Fe) in mg/L
Lead (Pb) in mg/L
Manganese (Mn) in mg/L
Mercury (Hg) in mg/L
Molybdenum (Mo) in mg/L
Selenium (Se) in mg/L
Uranium (U) in mg/L
Ammonia as N (N) in mg/L
Radium-226 (Ra-226) in pCi/L
3.5.1.4 State of Wyoming Requirements
The State of Wyoming's Land Quality Division of the Department of Environmental Quality
provides guidance for pre-mining water quality sampling in Appendix 1 to Guideline 8
(http://deq.state.wy.us/lqd/guidelns/Guideline8.pdf). Table 3-3 below lists these water quality
constituents.
3.5.1.5 State Ground Water Classification Systems
Several states have ground water classification systems that have been used in regulatory
restoration decisions when it has not proven possible to restore the ground water to pre-mining
conditions even after extensive remediation work. In some cases, the regulator has agreed that,
after extensive restoration, if the ground water meets the same usage classification, then the
restoration was approved even though specific analytes were not returned to pre-mining levels.
For example, Wyoming classifies ground water as follows [Wyoming Department of
Environmental Quality (WDEQ), Water Quality Rules and Regulations, Chapter 8 -
http://deq.state.wv.us/wqd/WQDrules/Chapter 08. pdfl:
• Class I - suitable for domestic use.
• Class II - suitable for agricultural use (where soil conditions and other factors are
adequate).
• Class III - suitable for livestock.
• Class Special (A) - suitable for fish and aquatic life.
• Class IV - suitable for industry (Class IV(A) - TDS < 10,000 mg/L; Class IV (B) - TDS
>10,000 mg/L).
Draft Technical Report
37
Revised Draft - November 26, 2012
-------
• Class V - associated with commercial deposits of hydrocarbons and/or other minerals or
considered a geothermal resource.
Table I of Chapter 8 of the WDEQ regulations specifies allowable concentrations for Class I, II,
and III constituents. In the case of ISR facilities, the water would likely be Class V. According to
the regulations, any discharge into the Class V ground water "shall not result in degradation or
pollution of the associated or other ground water and, at a minimum, shall be returned to a
condition and quality consistent with the pre-discharge use suitability of the water."
South Dakota water quality regulations specify that ground water containing 10,000 mg/L or less
of IDS is "classified as having the beneficial use of drinking water supplies, suitable for human
consumption" (South Dakota Administrative Rules 74:54:01:03 -
http://legis.state.sd.us/rules/DisplayRule.aspx7Rule=74:54:01:03). It is not clear how the state
addresses ground water with concentrations above 10,000 mg/L.
In Texas, under the Texas Risk Reduction Program rule, all ground water-bearing units affected
by, or reasonably anticipated to be affected by, chemicals of concern having concentrations at or
above residential ground water assessment levels must be characterized with regard to the
applicable ground water resource classification (TCEQ 2010). A "ground water-bearing unit" is
defined as a saturated geologic formation, group of formations, or part of a formation that has a
hydraulic conductivity equal to or greater than 1 x 10"5 centimeters per second (cm/sec). The
Texas Risk Reduction Program establishes three categories of ground water resources,
designated Classes 1, 2, and 3, based on a site-specific evaluation of the current use of the
ground water-bearing unit, as well as its potential use, as defined on the basis of natural water
quality and well yield. Only saturated geologic units with hydraulic conductivities of
K >1 x io~5 cm/sec meet the definition of a ground water-bearing unit.
The ground water class establishes the types of response measures that may be employed
(decontamination/removal versus control) and the ground water protective concentration level
(PCL). For Class 1 ground water resources, affected ground water must be removed and/or
decontaminated to the critical PCL; control options are not permitted. For affected Class 2 or
Class 3 ground water resources, affected ground water must be removed and/or decontaminated
to the critical PCL, unless a plume management zone is approved, or such remediation is
demonstrated to be technically impracticable, in which case a plume management zone is
required. A key factor in establishing the ground water class is the TDS:
• Class 1 - < 1,000 mg/L or < 3,000 mg/L and meets 40 CFR Part 141 NPDWR standards.
• Class 2-< 10,000 mg/L.
• Class 3-> 10,000 mg/L.
The PCLs for Class 3 water are 100 times the PCLs for Classes 1 and 2.
In Nebraska, according to Title 118, Chapter 2, Section 002, of the Department of Environmental
Quality Regulations (http://www.deq.state.ne.us/RuleandR.nsf/Pages/118-TOC).
Draft Technical Report 3 8 Revised Draft - November 26, 2012
-------
The ground water standards and ground water classifications shall apply to all
ground waters of the State with the following exceptions:
002.01 Within an aquifer or apart of an aquifer that has been exempted through
the Rules and Regulations of the Nebraska Oil and Gas Conservation
Commission or through the Nebraska Department of Environmental Quality's
[NDEQ] Title 122-Rules and Regulations for Underground Injection and
Mineral Production Wells. This exception will apply only for ground water
contaminants directly related to the activity requiring exemption. If the exemption
designation is removed, this exception will no longer apply.
Wells at uranium ISR facilities in Nebraska are designated as Class III Mineral Production
Wells. This is the same classification as used by EPA (see 40 CFR Parts 144-148).
While class of use considerations have been used in determining the acceptability, it should be
noted that a determination based on the "class of use" is not authorized in EPA uranium milling
or RCRA regulations as a factor for setting ACLs. Consequently, such state protections are not
consistent with the stricter UMTRCA protection requirements.
3.5.2 Summary of Species Potentially Required for Compliance Monitoring - Tiered
Approach
Table 3-3 summarizes parameter measurements required by EPA, NRC, the State of Texas, and
the State of Wyoming to characterize pre-operational baseline water quality. The table also lists
the actual parameters monitored at the proposed Dewey-Burdock ISR site in South Dakota.
Table 3-3. Comparison of Ground Water Parameter Measurements Established by
Various Regulators with Actual Field Measurements from Dewey-Burdock Site
Species
EPA
Regulations
NRC Standard
Review Plan
(NRC 2003)
Dewey-Burdock
(Powertech 2009)
TCEQ
WDEQ
Trace and Minor Elements
Aluminum
Antimony
Arsenic
Barium
Beryllium
Boron
Cadmium
Chromium
Cobalt
Copper
Fluoride
Iron
Lead
Manganese
Mercury
Molybdenum
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yese
Yes1
Yes1
Yese
Yes1
Yes1
Yes1
Yes1
Yes
Yes1
Yes1
Yes1
Yes1
Yes1
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes1
Yes
Yes
Draft Technical Report
39
Revised Draft - November 26, 2012
-------
Table 3-3. Comparison of Ground Water Parameter Measurements Established by
Various Regulators with Actual Field Measurements from Dewey-Burdock Site
Species
Nickel
Selenium
Silver
Strontium
Thallium
Tin
Uranium
Vanadium
Zinc
EPA
Regulations
Yes
Yes
Yes
Yes
Yes
Yes
Yes &V2O5
Yes
NRC Standard
Review Plan
(NRC 2003)
Yes
Yes
Yes
Yes
Yes
Yes
Dewey-Burdock
(Powertech 2009)
Yes1
Yes inc. SelV&SeVI1
Yes1
Yese
Yese
Yes1
Yes
Yes1
TCEQ
Yes
Yes
WDEQ
Yes
Yes
Yes
Yes
Common Constituents
Alkalinity
Ammonia
Bicarbonate
Calcium
Carbonate
Chloride
Magnesium
Nitrate
Nitrite
Potassium
Silica
Sodium
Sulfate
Sulfide
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Chemical and Physical Indicators
Anion/Cation Balance
Sodium Adsorption
Potential
Specific Conductivity
pH
Redox Potential
Total Dissolved Solids
Yes
Yes
Yesa
Yesa
Yesb
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Radiological Parameters
Gross Alpha
Gross Beta
Gross Gamma
Lead-210
Polonium-210
Radium-226
Radon-222
Radium-228
Thorium-230
Thorium-232
Beta + Gamma
Yes
Yes + Ra-228
Yes
Yesc
Yes
Yesd
See note d
Yes
Yes
Yes
Yes8
Yesg
Yes8
Yes
Yesg
Yes
Yes
Yes
Yes
Yes
Yes
a - field and laboratory determination; b - laboratory only; c - excluding radon, radium, and uranium; d - if site
initial sampling indicates the presence of Th-232, then Ra-228 should be considered in the baseline sampling or
an alternative may be proposed; e - total; f - dissolved and total; g - dissolved, suspended, and total.
Draft Technical Report
40
Revised Draft - November 26, 2012
-------
Hazardous constituents listed by EPA but not included in the NRC Standard Review Plan include
aluminum, antimony, beryllium, thallium, gross alpha, and beta plus gamma. However, as shown
in Column 4 of Table 3-3 for the proposed Dewey-Burdock site (which is undergoing technical
review by NRC), all of the hazardous constituents listed by EPA are included in the ground
water sampling protocols. In fact, the pre-operational sampling conducted by the Powertech
Uranium Corporation includes several species not listed by either EPA or NRC. Several of these
additional consitutents were added by the State of South Dakota. Attachment A provides
additional details on the Dewey-Burdock background sampling program.
The NRC Standard Review Plan includes boron, which EPA has determined not to be a
hazardous constituent, as noted above.
The sampling requirements of the TCEQ are more limited with respect to hazardous constituents
than those included in EPA regulations. Hazardous constituents not listed by TCEQ include
barium, beryllium, chromium, copper, nickel, silver, thallium, and vanadium.
In addition to hazardous constituents drawn from EPA regulations, a variety of other species
must be measured to assess chemical and geochemical reactions occurring during in situ leaching
and to provide early warning of potential wellfield excursions. Since the lixiviant used in the
leaching process is typically carbonate/bicarbonate solution enhanced with an oxidant
(e.g., oxygen or hydrogen peroxide), these species are typically included in ground water
sampling programs. Nonsorbing tracers that provide early warning of excursions from the
wellfield during operations are also sampled. As discussed below, these may include chloride,
total alkalinity, and conductivity. Since the mobilization of uranium involves oxidation-reduction
reactions, monitoring of constituents that provide indications of the redox state of the ground
water may also be required. These may include redox potential, iron (Fe+2/Fe+3), selenium
(Se+4/Se+6), As(III)/As(V), and S'2/SO4'2.
The U.S. Geological Survey (USGS) has suggested that dissolved oxygen should also be
monitored to better understand redox reactions occurring in the ground water (NRC 2007, p. 23).
To date, regulators have not included dissolved oxygen in the required suite of constituents to be
monitored in ground water.
The suite of constituents to be analyzed may vary over the life of a wellfield. A full suite of
samples including all the species listed by EPA and the additional species listed by NRC in its
Standard Review Plan (NRC 2003), shown above in Table 3-3, Columns 2 and 3, would provide
the basis for a sample suite to be used for establishing the wellfield baseline, Phase 1 in Figure
3-1. Some of the listed constituents may not be detected in the baseline ground water. However,
since there is a possibility that nondetectable concentrations of specific constituents could
become measurable because of the introduction of the lixiviant and geochemical changes within
the aquifer during mining, the same suite of samples should be used to characterize the ground
water at the end of mining (Phase 2 of Figure 3-1). If some of the sampled constituents are not
detected in Phase 1 or Phase 2, then consideration should be given to eliminating those
constituents from monitoring during Phases 3, 4, and 5 as defined in Figure 3-1. In addition, it
may be appropriate to eliminate other constituents from Phases 3, 4, and 5 monitoring if the
constituents are not deemed to provide significant incremental information about ground water
Draft Technical Report 41 Revised Draft - November 26, 2012
-------
quality. Species such as sodium, potassium, ammonia (unless it is added to the lixiviant), and
SiC>2 might fall into this category. However, if geochemical modeling is required, elimination of
these species may not be possible since an anion/cation balance will be needed. Temperature and
TDS may also be needed to confirm that the thermodynamic database for the geochemical model
is valid. The next section discusses these requirements in greater detail.
In summary, the initial list of analytes for monitoring during Phase 1 and Phase 2 would include
all of the inorganic species listed in the EPA regulations (Column 2 of Table 3-3), in addition to
any species listed in the NRC Standard Review Plan (Column 3 of Table 3-3) that are not on the
EPA list. For ground water monitoring during Phases 3-5, the initial list would be reduced by
eliminating those species not detected during Phase 1 and Phase 2. The monitoring list for
Phases 3-5 could be further condensed by eliminating certain nonhazardous species judged not
to significantly impact ground water quality or adversely impact the ability to perform
geochemical modeling.
Compiling listings of species to be determined for an ISR operation can be regarded as a
"tiering" approach in that the contaminants regulated by various regulatory authorities would
comprise the top tier in an analytical scheme. These species would include radionuclides and
other species with adverse health effects. They would always require measurement. Those
species used for other purposes, such as detecting excursions, determining water quality, and
geochemical modeling exercises would be the next "tier" in a sampling and analysis plan. Many
of these species would also be required to demonstrate compliance with regulatory requirements,
although not directly tied to assessing potential adverse health effects. An example would be a
species used as an indicator of excursions and as a measure restoration. Species on the lower tier
list may be dropped from a sampling and testing protocol at different stages of the ISR operation
if they are determined to have no role in controlling radionuclide and toxic metal concentrations
and contaminant migration. For any specific ISR operation, the listing of species for
measurement should be stated initially in the licenseing process and subject to revision on the
basis of data collected during operations (see Section 3.5.6 below for an example).
3.5.3 Well Construction and Low-Flow Sampling Methodologies
The construction of the production, injection or monitoring wells involves sealing the units
overlying the production zone with steel, fiberglass or PVC casing that is grouted in place (NRC
2009). A typical well completion is shown in Figure 3-8. Other potential types of well
construction are described in NMA (2007).
The type of grout, screen and casing materials are selected based upon the chemistry of the
lixiviant and ground water, the depth of the target interval and the expected injection pressures.
PVC or fiberglass casings are generally used in wells less than 1,000 ft deep, while deeper wells
or wells that will be subject to high pressures are generally constructed with steel or fiberglass
casing and well screens (NRC 2009). The potential that chemical reactions may take place
between the casing material and the dissolved constituents in the ground water also needs to be
considered in selecting well construction materials.
Draft Technical Report 42 Revised Draft - November 26, 2012
-------
As described in Puls and Barcelona (1996) and EPA (2010d), the water in the well casing may
not be representative of the aquifer, and therefore needs to be excluded from the ground water
samples. Wells are purged to some extent prior to sampling for the following reasons: the
presence of the air interface at the top of the water column resulting in an oxygen concentration
gradient with depth, leaching from or sorption to the casing or filter pack, chemical changes due
to clay seals or backfill, and potential surface water infiltration into the annular space of a poorly
constructed or deteriorating well.
Low-flow refers to the velocity with which water enters the pump intake and that is imparted to
the formation pore water in the immediate vicinity of the well screen. Water level drawdown
within the well provides the best indication of the stress imparted by a given flow-rate for a
given hydrological situation. The objective is to pump in a manner that minimizes vertical flow
within the borehole, and draws formation water through the well screen to the pump.
Ground water sampling at ISR facilities is typically based on low-flow methods stabilization of
the water quality parameters within pre-determined ranges based upon historical data (NRC
2006). Stabilization parameters generally include electrical conductance, pH, oxidation-reduction
potential, turbidity, dissolved oxygen and temperature (NRC 2006). Major differences between
typical procedures used in low-flow sampling at RCRA and CERCLA sites versus ISR facilities
are the pumping rates are generally much higher at ISR facilities (5 to 10 gpm versus 0.1 to
0.25 gpm), the screened intervals are generally greater than 25 feet (versus 10 feet) (NRC 2006),
and the well diameters for the ISR wells are much larger (4 to 6 inches versus 2 inches for
typical RCRA and CERCLA wells).
Draft Technical Report 43 Revised Draft - November 26, 2012
-------
Shallow Sands
Production Zone
Sandstone
Overlying Clay —
Cemenl Fill in Annular
Space
Fiberglass, PVC, or Steel
Casing 4" to 6" Dia.
Drill Hole 7" to
10" Dia.
100' Max.
Casing
Centralizers
Retrievable Well
Screen Liner
(Optional)
Casing Point
Underream Zone
(Optional]
Underlying Clay ZLtM=Jl
Source: NMA2007
Figure 3-8. Cross Section of a Typical Injection, Production, or Monitoring Well
Draft Technical Report
44
Revised Draft - November 26, 2012
-------
3.5.4 Species Required for Geochemical Modeling
To adequately comprehend the potential mobility of uranium in ground water, a sufficient degree
of site characterization has to be conducted to understand hydrogeologic and geochemical
conditions and also the extent and distribution of any contaminant plumes and sources. As
discussed in EPA 2002c, sample collection programs are designed around goals associated with
specific project objectives. Data quality objectives (DQOs) define the types, quality, and quantity
of data required by the various aspects of a project. Once the DQOs are developed, appropriate
sampling methodologies, analytical protocols, and specific methods may be considered and
selected. For instance, the DQOs may be different if project goals emphasize detection and
monitoring of contaminant concentrations rather than geochemical speciation modeling. The
types of data required to fulfill the conceptual model needs related to the geochemistry are
presented below.
3.5.4.1 A queous Chemistry
A simplified conceptualization of the major speciation reactions that geochemical modeling
attempts to simulate are shown in Figure 3-9. During active mining, an oxidant (lixiviant) is
injected that releases oxygen (©2) in order to solubilize the uranium from the uraninite (upper
mineral shown in Figure 3-9) into uranyl carbonate complexes (UO2CO3°), which are extracted
in solution to the surface. The injected oxidant also reacts with pyrite that releases dissolved iron
and subsequently precipitates some form of hydrous ferric oxide (FIFO). The HFO precipitate
that forms is a mixture of iron-bearing minerals that often includes microcrystalline goethite.
Uranyl ions derived from the uraninite will also be adsorbed to the surface of the goethite and as
uranyl surface complexes (Hfo_wOUO2+2) and removed from solution. This surface
complexation reaction could be particularly important in the restoration phase (Gard and
Mahoney2012).
Characterization of geochemical conditions within an aquifer requires careful collection of data
(e.g., major ions) necessary for input into geochemical models. The sensitivity of uranium to
redox conditions also requires that precautions be taken during sampling and analysis to ensure
that results obtained are representative of ground water chemistry. The following are considered
"core parameters for ground water" for predicting uranium concentrations (EPA 2002c) in
ground water:
• Temperature.
• Oxidation-reduction potential (corrected to redox), pH, alkalinity.
• Turbidity.
• Total and ferrous iron.
• Dissolved oxygen.
• Specific conductance.
• Dissolved organic carbon.
• Major ions (Ca2+, Mg2+, Na+, K+, Ci; S2; SO42", NO32', ammonium, phosphate).
• Aluminum, silica, manganese.
Collection of turbidity data is important for a variety of reasons, including its usefulness as a
sampling equilibration parameter using low-flow techniques where metals are contaminants of
Draft Technical Report 45 Revised Draft - November 26, 2012
-------
concern. Turbidity data can also be helpful in explaining data anomalies (e.g., high total iron
concentrations under oxidizing conditions) and evaluating the influence of particulates on metals
concentrations. The collection of major anion and cation data is important for proper evaluation
of the aqueous geochemistry of the system and for performance of accurate geochemical
modeling in support of a site conceptual model. Fluoride is important in some hydrogeologic
settings because of its high complexing capacity and thus, is needed to determine the adequacy
of ion balances for geochemical modeling purposes. Additional information on this topic is
covered in EPA 2002c.
Kinetics
Rate Controlled
Equilibrium Based
Complexation Reactions
Source: Card and Mahoney 2012
Figure 3-9. Major Chemical Reactions Involved in Uranium Recovery and Restoration
Specification of the appropriate methodology for the collection and analysis of the above listed
parameters is beyond the scope of this document; however, descriptions of techniques and
methods can be found in EPA 2002c, ASTM procedures (e.g. ASTM 2010), and the USGS
online field manuals. Certain parameters must be measured in the field because of stability
issues, and others are recommended to be field measurements (EPA 2002c). Parameters that
must be measured in the field include temperature, pH, Eh, dissolved oxygen, and turbidity. The
use of low-flow sampling is recommended as described in EPA 2010a. Specific conductance and
alkalinity are also recommended for field analysis (EPA 2002c). Redox indicators such as Fe(II),
S2", and H2 should be measured in the field rather than in the laboratory; EPA 2002c describes
special considerations with respect to their measurements.
3.5.4.2 Solid Phase Geochemistry
Solid phase characterization should be included for sites where uranium or other inorganic
contaminants in ground water are being evaluated. The solid phase should be tested to confirm
the form of the uranium associated with the mineralogy of the solid phase, and to determine its
Draft Technical Report
46
Revised Draft - November 26, 2012
-------
stability or mobilization potential (EPA 2002c). Examples of useful solid phase characterization
techniques include optical analysis of the mineralogy by a geologist/geochemist, x-ray
diffraction and petrographic studies, scanning electron microscopy, high resolution transmission
electron microscopy or microprobe analysis to determine the presence of elements and minerals,
cation exchange capacity, neutralization or acid-generation capacity, and extraction of
amorphous iron or aluminum coatings to assess the mass of sorptive material on mineral
surfaces. Depending on site-specific needs, a useful approach would be to use mineralogical
identification tools to confirm mineralogy, followed by chemical extraction to quantify mass of
the mineral. Organic coatings or other organic matter on the solid phase can exert significant
geochemical influence on ground water and should be characterized. Mineral phases that are
commonly used in geochemical modeling include uraninite, pyrite, calcite and goethite (Gard
and Mahoney 2012). A summary of site-specific parameters that can be used to characterize
solid-phase materials is presented in Table 3-4. The subsurface material bulk density should also
be measured, so that the chemical parameters listed in Table 3-4 can be reported on a mass basis.
Furthermore, a sufficient number of core samples should be collected to adequately characterize
the spatial variability of the lithology.
Table 3-4. Parameters for Solid Phase Characterization
Solid Phase Parameter
Oxidation Capacity
Reduction Capacity
Neutralization Capacity
Sorption Capacity
Ion Exchange Capacity
Sorbed Uranium
Solubilized Non-Target Trace Metals
Reagent Stability
Mineralogy (Bulk and Trace)
Non-Target Solid Phase Contaminant
Extractable Fe/Al/Mn
Extractable Sulfide
Total Organic Carbon
Total Inorganic Carbon
Reduction/Oxidation Reaction Rate
Microbial Activity or Physiology
Microbial Population
Parameter Description
Capacity of sediment to oxidize a reduced chemical (uranium or
introduced remedial reagent)
Capacity of rock matrix to reduce an oxidized chemical (uranium or
introduced remedial reagent)
Capacity of solid phase to buffer change in pH (acid or alkaline)
Total mass of uranium (and other trace metals) that can be partitioned to
solid phase by various mechanisms
Total mass of charged ions that can be partitioned to solid phase via an
electrostatic mechanism
Mass of uranium that is partitioned to solid phase
Mass of non-target trace metals associated with solid phase that may be
solubilized as a result of remediation
Identification of undesirable by-products produced during reaction
between remedial reagent and contaminant/ solid phase ore
Identity of mineral phases present in various size fractions of solid phase
Non-target constituents that may negatively interact with remedial
technology
Mass of Fe/Al/Mn extracted from solid phase using reagents designed to
attack specific mineral phases
Mass of sulfide extracted from solid phase using reagents designed to
attack specific sulfidic mineral fractions
Mass of carbon associated with organic solid phases in sediment
Mass of carbon associated with inorganic solid phases in the rock matrix
The rate at which solid phase will reduce/oxidize (consume)
oxidizing/reducing reagent
Characterization of the microbial processes/characteristics controlling
redox reactions by lowering activation energies
Identification of the species of microbes that inhabit contaminated solid
phase
Draft Technical Report
47
Revised Draft - November 26, 2012
-------
Table 3-4. Parameters for Solid Phase Characterization
Solid Phase Parameter
Aquifer Permeability
Bulk Density
Parameter Description
The ability of aquifer material to transmit ground water based on
measurement of porosity
Defined as the mass of many particles of the material divided by the total
volume they occupy
Source: Modified from EPA 2002c
Collection of solid phase samples in the field for later mineralogical analysis also requires
careful attention and handling in order to preserve conditions as they are in the aquifer. Minerals
that are stable under a reducing environment (such as in an aquifer) will be subject to alteration
when exposed to oxygen. Removal of pore water from the sample may also cause some
transformations. Significant changes in reactivity may result from changes in mineral structure
and surface area due to drying at even slightly elevated temperatures. Careful handling is
necessary for proper preservation of solid samples for subsequent characterization. Specification
of precise procedures for sampling and handling solid-phase samples is beyond the scope of this
report but is discussed in EPA 2002c. If it is necessary, however, that solid-phase
characterization be included as part of site investigations where ground water is contaminated
with uranium, then additional planning and expense will be involved compared to typical current
site characterization practices. For example, recommended procedures for sample cores obtained
in the field for mineralogical or other solid phase characterization are that they should be
immediately capped and frozen, placed in a cooler or freezer, and later thawed under an oxygen-
free or inert atmosphere (EPA 2002c). Procedures will vary depending on site-specific concerns
and the analyses being performed. It is reasonable to expect that characterization will be
necessary at locations within contaminated and uncontaminated areas of an aquifer with the
number of samples related to the degree of heterogeneity.
3.5.5 Species Required for Excursion Monitoring
While the focus of this report is on post-restoration monitoring, to ensure completeness of the
discussion of constituents requiring monitoring at ISRs, this section briefly discusses excursion
monitoring. Monitoring of chemical species that can serve as leading indicators in detecting
excursions before the arrival of hazardous constituents is an essential operational feature at ISR
facilities. NRC requires use of three such indicators and notes the following (NRC 2003):
The choice of excursion indicators is based on lixiviant content and ground-water
geochemistry. Ideal excursion indicators are measurable parameters that are
found in significantly higher concentrations during in situ leach operations than
in the natural waters. At most uranium in situ leach operations, chloride is an
excellent excursion indicator because it acts as a conservative tracer, it is easily
measured, and chloride concentrations are significantly increased during in situ
leaching. Conductivity, which is correlated to total dissolved solids, is also
considered to be a good excursion indicator (Staub\Qi al.], 1986; Deutsch [et al.],
1985). Total alkalinity (carbonate plus bicarbonate plus hydroxide) is an
excellent indicator in well fields where sodium bicarbonate or carbon dioxide is
used in the lixiviant. If conductivity is used to estimate total dissolved solids,
Draft Technical Report 48 Revised Draft - November 26, 2012
-------
measurements will be normalized to a reference temperature, usually 25 °C,
because of the temperature dependence of conductivity.
Calcium, sodium, andsulfate are usually found at significantly higher levels in in
situ solutions than in natural ground-water concentrations. The use of cations
(e.g., calcium 2+, sodium+) as excursion indicators is generally not appropriate
because they are subject to ion exchange with the host rock. The use ofsulfate
may give false alarms because of induced oxidation around a monitor well (Staub
[et al.], 1986; Deutsch [et al.], 1985)... Although water level changes in artesian
aquifers are quickly transmitted, water levels are generally not considered good
indicators, because water levels tend to have significant natural variability. The
applicant may choose to add a non-reactive, conservative tracer to in situ leach
solutions to act as an excursion indicator. The applicant is required to provide the
technical bases for the selection of excursion indicators.
3.5.6 Case History - Evolution of Constituent Monitoring List
Experience at the Crow Butte ISR site in Nebraska provides an example of how the list of
ground water analytes required to be monitored evolves during operations. The original NRC
license (SUA-1534, dated March 4, 1998) specified in Section 10.3B that the following species
be analyzed to establish the baseline (and by inference the parameters to be monitored after the
cessation of mining): alkalinity, barium, bicarbonate, boron, cadmium, calcium, carbonate,
chloride, chromium, copper, fluoride, iron, lead, magnesium, manganese, mercury, molybdenum,
nickel, nitrate, nitrite, pH, potassium, Ra-226, selenium, silica, sodium, specific conductivity,
sulfate, temperature, total dissolved solids, uranium, vanadium, and zinc. The highlighted items
here are parameters that were included in the license agreement but were not listed in the NRC
Standard Review Plan (NRC 2003) and shown in Table 3-3, Column 3 above. In addition, silver,
gross alpha, and gross beta are listed in the Standard Review Plan but not included in the Crow
Butte license.
Crow Butte subsequently requested that NRC remove alkalinity, bicarbonate, boron, carbonate,
chromium, nitrate, specific conductivity, and temperature from the list of monitored parameters,
and NRC granted the request as part of License Amendment 11, dated June 26, 2001
(Leach 2001).
Crow Butte's rationale for the requested changes was as follows:
• Alkalinity, bicarbonate, and carbonate can be evaluated as a single parameter - total
carbonate.9
• Boron was a constituent of concern because it could affect crop growth. Although small
amounts of boron were detected in pre- and post-mining ground water, there is virtually
no use of ground water for irrigation in the area, so boron would not affect surface plants.
• Post-extraction concentrations of chromium in the ground water were below the limit of
detection, thus negating the need for further analyses.
9 However, as noted above, data on individual species may be required if geochemical modeling is done.
Draft Technical Report 49 Revised Draft - November 26, 2012
-------
• Specific conductivity is a general indicator of the ionic concentration of the ground water.
However, the same information can be derived from measurement of TDS, which is a
required parameter.
• Measurements of nitrate showed that baseline and post-mining results were essentially
the same and were a factor of two lower than levels which might give rise to concerns
about health effects.
• Comparison of baseline and post-mining levels of silica showed that the silica
concentrations were essentially the same before and after leaching.
• Crow Butte argued that the uranium recovery process does not affect ground water
temperature, and NRC accepted this reasoning.
NRC allowed some constituents to be removed from the required monitoring list because they
found that the elements duplicated information provided by other parameters, because they
believed that the ISR process had little effect on baseline values, because the quantity present in
the ground water was below the limit, or because health effects would be insignificant. All of
these determinations were based on site-specific information, regulatory review, and license
amendment. EPA did not participate in these determinations.
3.5.7 Formal Approach to Acceptable Restoration
The primary goal of restoration is to return all analytes to the value determined by baseline
monitoring. According to NRC practice, this value may be the baseline mean value or some
higher value based on the statistical variability of the mean. However, it is well documented
throughout this report that restoration of all analytes to baseline conditions is a difficult task, and
baseline values may not be re-established for some species even after extensive remediation
measures. Addressing the behavior of those analytes that EPA has defined as hazardous should
be the focus during restoration (see Table 3-2).
For those analytes with EPA quantitative regulatory limits, as described in Table 3-2, any analyte
that is above baseline after restoration but below the prescribed regulatory limit should be
acceptable without further action in accordance with EPA requirements. As noted above, this
will require reconciliation of certain limits in §264.94 and the MCLs in 40 CFR Parts 141 and
143.
For analytes that exceed the nonmandatory MCLs, as described in 40 CFR Part 143 [Secondary
Drinking Water Regulations (SDWRs)], the operator is allowed to demonstrate to the NRC or its
Agreement States through geochemical and ground water modeling or other techniques that the
measured concentrations do not pose a significant risk to human health or the environment. For
example, it might be possible to demonstrate that the ground water transport time to the nearest
receptor well is on the order of thousands of years. In the case of high sulfate content, it might be
possible to demonstrate through geochemical modeling that the excess sulfate is important in
ensuring that radium is removed by precipitation. While iron has a nonmandatory limit of 0.3
mg/L, the existence of high soluble iron (Fe+ ) may be indicative of reducing conditions that tend
to suppress the solubility of uranium. However, according to 40 CFR 264.94, approval of such
higher levels requires a rigorous license modification approval process for individual ACLs for
Draft Technical Report 50 Revised Draft - November 26, 2012
-------
hazardous substances in the ground water. To date, this formal approval practice has not been
carried out by NRC for ISR licenses.
For analytes not defined as hazardous by EPA (see Table 3-2), but monitored to assess how
leaching affects the ground water, it may be possible for a licensee through geochemical
modeling to demonstrate to the Regulatory Authority that restoration levels above baseline are
acceptable. This would involve analytes such as alkalinity, carbonate, bicarbonate, calcium,
potassium, magnesium, silica, and specific conductivity. For example, such modeling might
demonstrate that, at equilibrium, the calcium and the carbonate would react to form additional
calcite, in turn reducing the potential for solubilizing uranium as a carbonate complex, or the
modeling might show that, at the measured carbonate levels, uranium carbonate complexes
would not form. Still another possible alternative is that, in solutions with high magnesium
concentrations, the complex MgUO2( €03)3 " can form (Dong and Brooks 2008).
Table 3-5 summarizes the various analytes that may require monitoring for the following
reasons:
• May be required based on EPA regulations
• May be used for excursion monitoring
• May be required to assess the chemical state of the lixiviant during operations and
restoration
• May be needed for geochemical modeling
• May be required by state regulations in addition to EPA requirements
Table 3-5. Basis for Inclusion of Various Analytes in ISR
Ground Water Monitoring Program
Species
EPA
Regulations
Used to
Assess State
of Lixiviant
Needed for
Geochemical
Modeling
Required by State
Regulations but
Not by EPAa
Used for
Excursion
Monitoring
Trace and Minor Elements
Aluminum
Antimony
Arsenic
Barium
Beryllium
Boron
Cadmium
Chromium
Cobalt
Copper
Fluoride
Iron
Lead
Manganese
Mercury
Molybdenum
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes (total & Fe+2)
Yes
Yes
Draft Technical Report
51
Revised Draft - November 26, 2012
-------
Table 3-5. Basis for Inclusion of Various Analytes in ISR
Ground Water Monitoring Program
Species
Nickel
Selenium
Silver
Strontium
Thallium
Tin
Uranium
Vanadium
Zinc
EPA
Regulations
Yes
Yes
Yes
Yes
Yes
Yes
Yes &V2O5
Yes
Used to
Assess State
of Lixiviant
Yes
Yes
Needed for
Geochemical
Modeling
Yes
Required by State
Regulations but
Not by EPAa
Used for
Excursion
Monitoring
Common Constituents
Alkalinity
Ammonia
Bicarbonate
Calcium
Carbonate
Chloride
Magnesium
Nitrate
Nitrite
Phosphate
Potassium
Silica
Sodium
Sulfate
Sulfide
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes (NH4+)
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Chemical and Physical Indicators
Anion/Cation Balance
Sodium Adsorption
Potential
Specific Conductivity
pH
Redox Potential
Total Dissolved Solids
Temperature
Turbidity
Dissolved Oxygen
Dissolved Organic and
Inorganic Carbon
Eh
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Radiological Parameters
Gross Alpha
Gross Beta
Gross Gamma
Lead-210
Polonium-210
Radium-226
Radon-222
Yes
Yes + Ra-
228
Yes
Yes
Yes
Yes
Draft Technical Report
52
Revised Draft - November 26, 2012
-------
Table 3-5. Basis for Inclusion of Various Analytes in ISR
Ground Water Monitoring Program
Species
Radium-228
Thorium-230
Thorium-232
Beta + Gamma
EPA
Regulations
Yes
Used to
Assess State
of Lixiviant
Needed for
Geochemical
Modeling
Yes
Required by State
Regulations but
Not by EPAa
Yes
Used for
Excursion
Monitoring
a - Texas and/or Wyoming
It should be emphasized that Table 3-5 provides generic lists that will need modification on a
site-specific basis. This is particularly true of the geochemical modeling parameters, which will
vary depending on the type(s) of modeling that is planned or required.
Draft Technical Report
53
Revised Draft - November 26, 2012
-------
4.0 TECHNICAL CONSIDERATIONS FOR ISR GROUND WATER MONITORING
Monitoring wells within an in-situ mining area and site vicinity serve vital functions necessary
for efficient uranium recovery with minimal adverse environmental impacts. Proper monitor well
placement and data collection from these wells ensure that the aquifer constituents are detected
and then restored to pre-mining levels. Without adequate placement of monitoring wells and
proper data collection, which includes consideration of sample frequency and sampling time
frame, mine operators and regulators (1) may not detect excursions of lixiviant outside the
mining area during operations, and (2) may not be able to confidently determine whether the
impacted aquifer needs further restoration or has been restored to its pre-mining state or to
predetermined conditions specified by regulators.
This section focuses on technical considerations for ground water monitoring through all
operational phases of an ISR facility. Because the monitoring goals and practices depend on the
characteristics of the ore body, this section begins with a discussion of geographic, geologic, and
chemical characteristics typical of uranium deposits suitable for leaching.
4.1 Uranium Geology
The geographic areas that are considered to be potential resources of uranium in the United
States are shown in Figure 4-1. The principal regions of uranium recovery by ISR are the
Wyoming basin, the Colorado Plateau, and the Gulf Coastal Plain of Texas (Figure 4-2). The
southern Black Hills in South Dakota and northeast Colorado/western Nebraska within the Great
Plains region also contain sedimentary uranium deposits amenable to ISR. Furthermore,
exploration is ongoing elsewhere in the U.S. that could expand the extent of production to other
states, such as Michigan or Alaska, and to types of uranium deposits (other than roll-front) that
could be extracted through ISR methods.
(Source: DOE 1980)
Figure 4-1. Uranium Resource Areas of the United States
Draft Technical Report
54
Revised Draft - November 26, 2012
-------
Christensen Ranch . .
Moore Ranch # Smilh Ranch
(Source: NMA 2007)
Figure 4-2. Pending, Licensed, and Active ISL Operations
4.1.1 Formation of Uranium Containing Ore Deposits
Based on the shape of the ore body and relationship to the deposit!onal or structural environment,
sandstone uranium deposits can be subdivided into three types (these may be gradational into
each other): tabular, roll-front, and remnant-primary sandstone (Hou 2004). Tabular and roll-
front mineralized bodies form along the contact of sand and intercalated clay horizons and at
paleochannel margins (Figure 4-3), while remnant-primary sandstone deposits may occur in
sandstones adjacent or vertically stacked along a permeable fault zone (Figure 4-4). Precipitation
of uranium minerals in most tabular deposits is thought to begin shortly after sedimentation and
burial. Mineral detritus and rock fragments derived from weathered bedrock are deposited, along
with channel sediments. The uranium is leached under oxidizing and slightly acidic conditions
and is mobilized in ground water moving through the sediments, with mineralization commonly
accompanying diagenesis of the sediments.
Draft Technical Report
55
Revised Draft - November 26, 2012
-------
0 20 40 60 80 100 120 140
(Source: Hou 2004)
Figure 4-3. Three-Dimensional Depiction of Uranium Ore Deposited in Paleochannels
Fracture zone
BRUSHY BASIN MEMBER
RECAPTURE MEMBER
Source: NRC2006
Figure 4-4.
Schematic Diagrams of the Different Geometries for Tabular, Roll-front,
Fault Displaced, and Remnant Ore
Draft Technical Report
56
Revised Draft - November 26, 2012
-------
In Nebraska, New Mexico, Texas and Wyoming, the general geology and formation of roll-front
deposits is nearly identical with the host rock being relatively near-surface sandstone that was
deposited in fluvial or lacustrian environments (NMA 2007). Aquifers and lithologic units
underlying topographically high areas, such as ridges, are recharged by rainfall that has elevated
oxygen and carbonic acid, which solubilizes low concentrations of uranium (and other metals)
from the soils and rocks. The mobilized uranium moves deeper into the formation into areas
where the ground water is being reduced by organic carbon, carbonized fossil wood, pyrite
and/or hydrogen sulfide gas (URI 2006; Devoto 1978). As the oxidized ground water and
dissolved uranium move into the reduced zone, the uranium is precipitated in its reduced mineral
form and removed from solution (Figure 4-5). Therefore, the roll is located at the interface
between the reduced and oxidized facies in the paleoaquifer (Devoto 1978).
Infiltration
Water Table
Appro*. 0.05-0.257
-------
unit because of spatial differences in depositional environments (URI2006). An excellent
discussion of the geochemical processes controlling the mobility of uranium, radium and
associated metals is presented in Demuth and Schramke (2006).
Impermeable (ConfiningVZon
Unaltered
(Reduced)
Sandstone
Hcmatitic Core
Groundwater Flow
I in permeable' (Confining) Zone
Hcmatitic Core Alteration Envelope Ore-Stage Uranium Ore-Stage Pyritc Reduced Sandstone
Hematite
Magnetite
Siderite
Sul
-------
The depth of the ore body varies significantly from site to site. Ore bodies in the Texas Gulf
Coastal Plain tend to be shallow-lying, while those in the Wyoming Basin show considerable
variability, as illustrated by the following:
• Burns/Moser TX (Lower Oakville) - 230 to 280 ft (http://www.wise-
uranium. org/udusail. html#KINGS V).
• Clay West TX (Lower Oakville) - 250 to 400 ft (http://www.wise-
uranium. org/udusail. html#KINGS V).
• Crow Butte (Basal Chadron) - 400 to 800 ft (NRC 1998).
• Dewey-Burdock (Lakota and Fall River) - 500 to 800 ft (Dewey); 300 to 500 ft
(Burdock) (SRK 2011).
• Irigaray WY - 100 to 300 ft (Irigary 2004).
• Palangana TX (Goliad Formation) - 230 to 390 ft (http://www.wise-
uranium.org/udusail.html#PALANGANA).
• Smith Ranch/Highlands, A-Wellfield WY (20-Sand) - 530 ft avg (Kearney 2004).
• West Cole TX (Soledad member of Catahoula Formation) - 225 to 270 ft
(http://www.wise-uranium.org/udusail.htmltfWESTCOLE).
• Zamzow TX (Oakville sand) - 35 to 225 ft (http://www.wise-
uranium. org/udusail. html#Z AMZOW).
It is expected that deeper lying ore bodies will be less susceptible to seasonal variations in
ground water chemistry associated with areal recharge than shallower ore bodies.
4.2 Aquifer Exemption Requirement
An aquifer exemption is required if the proposed injection zone is an underground source of
drinking water. Applicants submit a Petition for Aquifer Exemption to the state for review.
Although the state reviews the aquifer exemption petition, it cannot grant the final aquifer
exemption. After the state makes a decision on the aquifer exemption petition and it is duly
public noticed, the state makes a recommendation of the decision to EPA. EPA makes the final
decision regarding aquifer exemptions.
As part of the aquifer exemption Petition, the Applicant must evaluate potential impacts on the
water resources in the vicinity of the proposed mine. This evaluation generally involves water
quality and use surveys within at least a 2-mile radius of the site. Water level measurements and
ground water samples collected during this phase of the application process often provide some
of the earliest baseline data.
4.3 Establishing Baseline Conditions
Before initiating leaching activities, knowledge of the aquifer baseline characteristics is needed
to help determine restoration goals for the post-mining phase. Pre-mining monitoring and testing
wells are installed to collect data that define the ground water flow regime through the extraction
Draft Technical Report 59 Revised Draft - November 26, 2012
-------
zone and surrounding areas and determine the chemical characteristics of the ground water.
Monitoring wells should be installed at well locations up gradient, down gradient, and tangential
to the proposed ISR field, as well as within the "ore-zone." Well placement should be designed
to measure all potential "escape" pathways for introduced constituents and mobilized metals, as
well as to provide data to determine the choice and effectiveness of aquifer restoration actions.
Selected up gradient and down gradient wells outside the wellfield should be continuously
monitored throughout all phases of the operation illustrated in Figure 3-1. This will provide data
that may assist in interpreting changes in ground water chemistry within the wellfield during
steady state and long-term stability monitoring and possibly to detect seasonal variations in
ground water chemistry in shallow deposits.
The design of the monitoring network is largely a site-specific decision predicated on a thorough
knowledge of the ground water flow regime and the effects of the injection and withdrawal rates
on the flow system behavior. A system of wells should be emplaced to monitor the horizontal
and vertical ground water velocity and flow paths, the ground water chemical conditions, and the
potential for hazardous constituents to migrate beyond the ISR mine field, both within the mined
aquifer and through transmission of contamination to overlying and underlying aquifers. These
areas beyond the ISR may experience contamination from the mined area.
The following components and parameters need to be considered in establishing baseline site
characteristics (see Section 3.5, and particularly Table 3-3, for more details on analytes to be
monitored in ground water):
(1) Hydrogeochemical Conditions - Eh (including redox-sensitive couples), dissolved
oxygen, pH, major ions, TDS, carbonate alkalinity, pCC>2, radioactive constituents,
colloids, organic constituents, hydrogen sulfide, trace elements (to be compared against
post-restoration measurements).
(2) Concentrations of those constituents listed in 40 CFR Part 192 - arsenic, barium,
cadmium, chromium, lead, mercury, molybdenum, nickel, Ra-226 and Ra-228, selenium,
silver, uranium, etc.
(3) Uranium Ore Deposit Types and Oxidation States - The site-specific and varied diagenic
processes that formed the uranium deposits will determine how ISR operations will affect
baseline conditions and which restoration approach is likely to be most effective.
Knowledge of these processes can be used as a framework in estimating the time needed
for the aquifer to reach baseline conditions once post-mining restoration and monitoring
are initiated.
(4) Hydrogeologic Setting - Pre-mining ground water velocities (unstressed), flow paths, and
solute transport time frames. A reliable and defensible characterization survey of the ISR
site requires thorough core and water sampling from all monitoring wells and exploration
boreholes. Sufficient data must be collected before the mining activity commences to
understand when baseline levels have been reached after mining. Aquifer pump/stress
tests and core sample analysis will determine aquifer characteristics within and
surrounding the ore body and be used to determine:
a. Host rock and ore zone permeability, porosity, storativity, and thickness
Draft Technical Report 60 Revised Draft - November 26, 2012
-------
b. Whether more monitoring wells are needed for postclosure activities and to assess
the time frame of postclosure monitoring
c. Time frame estimates after mining has ceased, in order for the system to reach
pre-ISR conditions
d. Recharge/discharge points
e. Impermeable layers above and below the ore zone
f Proximity to ground water barriers
g. Proximity to surface water bodies - natural or manmade
Sampling the ground water may require special sample collection techniques, depending on the
chemical constituents of concern. For major ions and some other chemical species, sampling
may be relatively simple, in that these species are not susceptible to change upon exposure to
atmospheric conditions. For species that are susceptible to re-equilibration in response to
atmospheric conditions, particularly redox-sensitive species and the carbonate-bicarbonate
system, water sampling may require that the sampled interval be "packed off within the well
and water samples taken in containers, which were placed within the sealed intervals prior to the
"packing-off' and left to equilibrate in the flowing ground water for a period of time before
removal. Redox-sensitive couples typically examined include ferrous (II)/ferric (III) iron, and the
arsenic (III) /arsenic (V) couple. In addition to dissolved oxygen levels, these couples can
produce important characterization of the redox conditions in the production zone before, during,
and after the leaching process and can also be important in determining the effectiveness of
various aquifer restoration processes.
In addition, uranium speciation is strongly affected by pH and carbonate concentrations in the
ground water, which, in turn, are a function of the pCC>2 in the ground water. Exposure of the
ground water sample to the atmosphere can result in the escape of CC>2 and re-equilibration of
the uranium-carbonate system due to the out-gassing. Then the uranium concentrations in the
reequilibrated water would not reflect the actual speciation in situ, and, consequently, could
result in misleading calculations of uranium speciation and solubility constraints in the
subsurface waters. Because of these effects and their relative importance to characterizing the
in-situ ground water chemistry, monitoring water chemistry in and around the "ore body" may
well require differing sampling methods.
Figure 4-7 illustrates the spatial variability that exists between the ore zone and its surroundings.
The figure summarizes results for pre-mining sampling of excursion monitor wells (designated
as EM) and production zone wells (EMP) for Wellfield H-E at the Highland uranium project in
Wyoming (Hoy 2006). Within the production zone, most of the baseline wells showed elevated
concentrations of uranium and radium as compared to the excursion monitor wells located a few
hundred feet from the production zone wells. Within the production zone, radium (Ra-226) was
generally >100 picocuries per liter (pCi/L), while at the excursion monitor wells, Ra-226 was
generally <5 pCi/L. Uranium was generally >0.03 mg/L in production zone baseline wells, while
it was always <0.030 mg/L in the baseline monitor wells. The median uranium concentration was
about 0.04 mg/L within the production zone and about 0.015 mg/L in the monitor ring. It is
possible that different geochemical conditions within the ore zone, as compared to the host rock,
Draft Technical Report 61 Revised Draft - November 26, 2012
-------
precluded the movement of uranium and radium down gradient. (See also Figure 4-4 and Figure
4-6.)
The wide variablitity and sharp concentration gradients of background uranium concentrations
observed at the Rosalita ISR project are shown in Figure 4-8 (URI2010). The isolated areas of
elevated uranium concentrations within the ore body present a challenge to establishing
statistically representative background concentrations as described below in Section 4.2.1.
Recent studies (Hall 2009; NRDC 2012), as well as administrative hearings for the Goliad ISR
aquifer exemption that took place in 2010, have focused attention on a lack of guidance on
statistical establishment of background levels for hazardous constituents in ground water, and
selective use of well data, which could distort the chosen background levels for restoration. Use
of statistics in setting these numbers is discussed further in Section 7 of this report.
Draft Technical Report 62 Revised Draft - November 26, 2012
-------
Base Map Source:
EM-28 XCM-30
Adapted from Plate 1, Volume 21 (1/2), Permit 603. EM-27 LLL xCM_2g
Note:
Uranium & radium should be dissolved
concentrations, & radium is Ra-226 only.
CIVI-ZD 1 — 1 — 1 1 ,,-" ^ ~--,v
rm .„-*'' ^'cM-28
EM-25 .®''
CLXL®-" EMP.16 xCM-27
Explanation
Monitor Well Rings
Welffield"E" -Baseline
EM-24 ,.•' MM EM-28 Well Location &Number
EM
I I .l@-' EMP-1 5 ^ xCM-26
-23 -"' '^x'
WeIlfield"C" -BaselineWell
CCE] as"'"' * ™, ic CM-19 Locations Number
EM-22
.-• EMP-13
TT1 *^..nH, R73W.22123 >< CM-19
EMP-1 4 . CM.24
/'® EMP-10 V^ I I •
EM-21 / [
EM-20 _a' rffn9
EM-1 9 ,..•&"" EMP-7
rm .'-•" EMP-5 1 i M
— — CEf ' 1 1 1 ^ ' '
..•'® EMP-3 LLLk
EM-i 8 ,/ n~n X
I I I I 9 EMP-1 * *
I I I I ^ % EMP-4
EMP-2 LJ-L-1
0 rm
\ EM-1 7
1 1 I I I
EM-1 6 """---^
III EM-15 """---.._
EM-14+ """---.._
L-LU EM-1 3
cm
TT1 * EMP-12 ,,., „ CM 21 x CM-20
V * "CICB EMp-18 x . 23 |
Production Zone
Pattern Area
*B Baseline Monitor Well
"'* * V EMP-1 1 L-LJJ&j i EMP-28 Location & Number
'-'•'••W'i PMPIQ /•'•> CM-22 ft, EM-1
*^EMp-8 Epf^ ?cnn
EMP-6 III " EMP-32 : ^^ Baseline Water Quality
EMP-1 7 EMP-21 "'*' ffi EM-3
Ml MM * EMP-31 ; [
,W n^
EMP-22 ^ ^ EM-4
Radium (| | <5, >5 I I ^10?
>10D<100, D>100pCi/l)
Uranium (D^O.03, D>0.03 mg/1)
I I I I * EMP-30 ^ ,-'' Mil TDS(n<500,>500D<1200mg/l)
EMP-23 ,,.-'r" *EM-5
Ml ^ ^EMP-29 S* ||| — i Ml I — r~Tl Well location found, but no
EMP-24 Mil <^FM 7 water quality analyses found.
EMP-25 ^ EMP-28
U^ * "Wt" ffiEM-8
§ 'VEMP-26 EMP-27 /' MTT1
' / j\r
_----"* EM-9
---®" rm
EM-12^-- €>-"" EM-10
rm EM-H Mm
nm
0 600 12QO
^^•^^a^^z^^
Approximate Scale (feet)
Source: Hoy 2006
Figure 4-7. Well and Production Zone Locations and Baseline Concentrations of TDS, Uranium, and Radium -
Wellfield H-E
Draft Technical Report
63
Revised Draft - November 26, 2012
-------
URI's Rosita ISR Project
Baseline Uranium (U) [ppb]
[Drinking Water Standard: 30 ppb]
(Source: URI 2010)
Figure 4-8. Baseline Uranium Concentrations at the Rosita ISR Facility
4.3.1 Variability in Baseline Measurements
The two principal sources of variability in ground water quality data are "natural" variability and
variability that results from the network design and operation. The components of natural
variability arise from temporal or spatial variability related to hydrologic processes such as
pumpage, recharge, or discharge, as well as influences of these processes on the release and
distribution of chemical constituents from a variety of chemical sources. The sources may be
natural mineral assemblages, precipitation, and percolation through the unsaturated zone, in
addition to numerous point (e.g., ISR injection wells) and nonpoint sources (e.g., agricultural
application) of chemical contaminants. In general, natural sources of variability cannot be
controlled, although they may be quantified through an effective monitoring network design.
Variability in water quality data may also arise from the sampling and analytical components of
the monitoring network design. Sampling variability includes variations due to the selection of
the locations and construction of sampling points in space, sampling frequency, well purging,
and the execution of the sampling protocol. The sampling protocol consists of the procedures
used to collect, handle, preserve, and transport water samples to the analytical laboratory.
Several authors have evaluated elements of the sampling protocol for their relative contributions
to variability or errors in water quality data (Barcelona et al. 1983, 1984, 1985a, 1985b; Garske
and Schock 1986; Barcelona and Helfrich 1986).
Analytical variability in water quality data arises principally from the errors involved in
analytical methods and the subsequent data processing steps. These errors can be controlled once
Draft Technical Report
64
Revised Draft - November 26, 2012
-------
suitable water quality indicators or chemical constituents have been selected and a thorough data
quality assurance/quality control program has been designed and executed.
Optimization of sampling frequency in ground water quality monitoring networks should provide
sufficient sensitivity for chemical constituent detection and adequate characterization of average
chemical conditions. This should be accomplished with a minimum number of sampling times,
but this sampling should provide important information for determining the adequacy of
restoration and necessitates a high degree of confidence. This may then require more lengthy
sampling periods.
The hydrology of the ground water system will influence the impact of the sources of variability
mentioned above. Although aquifer hydraulic properties may not vary significantly at a single
measurement point over time, spatial variability may be substantial. This is a very active area of
research with application to monitoring network design (Christensen and Doherty 2008, Meyer
et al. 2007, Cooley 2004, Doherty and Hunt 2009). Results of such research will be important for
determining the efficacy of restoration efforts and regulatory decisionmaking about the long-
term stability of the restored wellfield.
Temporal and spatial variations in ground water elevation may affect ground water flow rate and
the direction of movement. Such changes may influence the quality of the ground water in the
vicinity of a sampled well by directing water from a different up gradient area or changing the
velocity with which dissolved constituents move along a flow path. Examples abound in the
literature detailing ground water response (i.e., elevation change) to a wide variety of influences.
In addition to seasonal fluctuations produced in response to short-term (i.e., months to 1 year)
events, ground water levels also reflect changes in long-term (i.e., years to decades) conditions.
Natural and artificial (man-induced) influences can cause changes in ground water elevation,
including natural (e.g., rainfall and snowmelt) and artificial recharge (e.g., pipe leaks, injection
wells) and natural (e.g., evapotranspiration) and artificial discharge (e.g., pumping). These
variations may be important in situations where multiple ore zones are mined sequentially, and
decisions on when and where restoration efforts are to be carried out and regulatory decisions on
these efforts must be made.
Ground water quality monitoring networks are designed for a number of purposes, including
ambient resource studies, contaminant detection and assessment, contaminant source evaluation,
and research investigations. The effective design of virtually any such network, regardless of
purpose, depends on knowledge of the hydrogeologic system of interest, an indication of the
presumed contaminants or preferred water quality indicators, and an assessment of the relative
contributions of sources of variability. These aspects of monitoring network design have been
addressed in the literature (Todd et al. 1976, Sanders et al. 1983, Moss et al. 1978, Liggett 1984,
Liggett 1985, Gillham et al. 1983). The common recommendation in these works is that
background information must be supplemented with the results of a preliminary sampling to
progressively refine the network design to account for error and variability in the chemical
results.
Variability in the analytical results for particular ground water chemical constituents may arise
from "natural" causes such as nonhomogeneous spatial distributions of the constituents and
temporal variability in recharge. Variability may also arise due to network design-related factors
Draft Technical Report 65 Revised Draft - November 26, 2012
-------
such as well design, sampling devices, and sampling protocols. The apparent sources of
variability in water quality data, which are often attributed to natural (i.e., temporal and spatial)
effects, include hydrologic transience, the fluctuations in contaminant source strength and
composition and the interactions between reactive chemical, biochemical, and mineral
constituents in recharge water and ground water. A detailed understanding of the
interdependence of hydrologic, biological, and chemical processes in the subsurface is limited.
However, it may not be necessary to fully understand the relationship between these processes,
contaminant sources, and the resultant chemical distributions in order to establish representative
background concentrations, monitor potential contaminant releases (e.g., excursions), and predict
remediation times.
The temporal and spatial variability, which is observed in water quality results over time at
discrete monitoring points, is the result of the processes noted in the preceding discussion, as
well as the sample collection and measurement errors inherent to network design and operation.
This variability, or "noise," in the data encompasses the stochastic distribution of possible values
for particular chemical constituents and the effects of both determinate (i.e., systematic) and
indeterminate (i.e., random) error. Determinate error can be measured as inaccuracy or bias if the
"true value" is known. Indeterminate error can be estimated as imprecision or irreproducibility if
enough replicate determinations can be made to faithfully estimate the mean or the "true" value.
In practice, determinate errors can be estimated and controlled only by careful quality
assurance/quality control measures exercised over appropriate sampling and analytical
procedures because the true value in environmental distributions is unknown, and some
disturbance of the subsurface is inevitable in ground water quality work. Identifying and
controlling these design-related errors are described in several sources (Barcelona et al. 1983,
Barcelona et al. 1985a, Barcelona 1988, Barcelona et al. 1986, Barcelona and Gibb 1986, Puls
and Barcelona 1996, and EPA 2010d).
Statistical measures of short-term temporal variability include seasonal effects
(e.g., consequences of recharge or temperature effects), which can be assigned to the seasons of
the year, periodic effects (e.g., consequences of anthropogenic contaminant sources or pumping
effects), and serial correlation or dependence effects, which tend to make data points following
maxima or minima in temporal data series higher or lower, respectively, than one would attribute
to random processes alone. Trends in data, on the other hand, are long-term variations compared
to those that may occur within a hydrologic year (Porter and Trautman 1984). This
categorization of temporal effects is somewhat artificial in that the combination of seasonal,
periodic, or correlative components may result in a water quality time series that cannot be
differentiated quantitatively. Because of this factor, the identification of short- or long-term
trends in water quality is conditional on some knowledge of the proximity of the sampling
location to the location and time of chemical release, as well as the statistical characteristics of
ground water quality variables.
Statistical measures of temporal variability have been reviewed by Loftis et al. (1986),
Montgomery et al. (1987), and Harris et al. (1987). They cite numerous examples of both
short- and long-term temporal variability, which supplement the earlier reviews of Porter and
Trautman (1984) and Colchin et al. (1978). The Air Force Center for Environmental Excellence
(AFCEE) has also developed two computer software packages that employ sophisticated
Draft Technical Report 66 Revised Draft - November 26, 2012
-------
geostatistical/ statistical analysis algorithms to facilitate the optimization and termination of
current long-term monitoring (LTM) programs (Aziz et al. 2006). The Monitoring and
Remediation Optimization System (MAROS) software assists in optimizing monitoring
networks both spatially and temporally.
A recent development in MAROS is the addition of the Modified Cost-Effective Sampling
(CES) Method, which is designed to set the sampling frequency for a well based on the analysis
of time series concentration data at each sampling location, considering both recent and long-
term trends of the concentration data. In contrast to the spatially based Delaunay Method
originally used in the MAROS sample location optimization, the Modified CES Method is based
on temporal analysis. Its use, combined with the Delaunay Method and trend analyses, leads to a
comprehensive process of sampling optimization. The second program developed by AFCEE is
the geostatistical temporal/spatial (GTS) optimization algorithm, which is a robust and powerful
computer application designed for use by mid-level geostatisticians. GTS software can optimize
individual input data sets for LTM networks containing analytical data from more than one
aquifer or hydrogeologic unit (MAROS can optimize only a single hydrogeologic unit).
Additional technical information, user's manuals, and executable computer code for MAROS
and GTS can be obtained from the AFCEE website. Section 7.0 provides a more detailed
discussion of the statistical approaches to address both spatial and temporal variability.
A sufficiently high sampling frequency (i.e., quarterly) is required to statistically distinguish
seasonal effects from those of serial dependence or autocorrelation (Loftis et al. 1986). In many
instances, limited ground water quality data sets, combined with quality variables that are
frequently not normally distributed, constrain the use of simple parametric statistical tests of
significance to compare means or identify trends (Montgomery et al. 1987).
The adoption of a minimum sampling frequency of "quarterly" can be useful during the
characterization phase of ground water monitoring to evaluate seasonality, rate of change and
variability, especially for fast-moving plumes. However, for more stable plumes, the default
adoption of quarterly monitoring may be determined to be unnecessary if long-term monitoring
can demonstrate no harm by utilizing longer intervals. Conversly, if variability in the ground
water compositions is relatively high and the concentrations of the analyte of concern over time
is changing slowly, more frequent sampling may be necessary to define the presence or absence
of trends with a reasonably high degree of confidence (see Sections 7.7 and 7.8 for an example
of these situations). The use of sampling intervals ranging from quarterly, semiannual, annual, to
biennial levels is very common in long-term ground water monitoring (Air Force 1997; NFESC
2000).
Spatial and temporal variability in ground water quality may affect the sensitivity of contaminant
detection and the estimation of mean chemical concentrations. To some extent, spatial chemical
data collected at discrete points along a horizontal flow path may be quite similar to data
collected over time at a single point in the path. This supposition depends, of course, on a
number of factors related to hydrologic conditions, as well as the nature of the chemical source,
reactivity, and mobility constraints. The substitution of spatially intense samples for use in
temporal variability studies could be applied to studies of ambient concentrations of conservative
chemical species for regional assessments in unique hydrologic situations.
Draft Technical Report 67 Revised Draft - November 26, 2012
-------
4.4 Extraction Operations Phase
During the ISR mining operations phase, wells are placed in the active ISR treatment zone,
fringe zone (wells at the perimeter of the ISR mine), and outside the affected areas (Figure 4-9).
The recovery wells (shown as red squares in Figure 4-10) are pumped at a slightly higher rate
than what is reintroduced into the aquifer through the injection wells (i.e., blue squares). The
injection creates ground water mounds around the injection wells (e.g., 98-foot contour) and the
pumping forms ground water depressions around the pumping wells (i.e., multiple concentric
circles of lower elevation). The net effect of this pumping/injection is the formation of capture
zones (Figure 4-11) in which the water/lixivant introduced through the injection wells (blue dots
in Figure 4-11) flows to the pumping wells (red squares in Figure 4-11) and is withdrawn.
The functions of a monitoring system during the extraction phase include:
(1) Monitoring the extraction process to determine uranium recovery rates within the mining
zone.
(2) Assessing the mass-balance of the lixiviant fluids.
(3) Monitoring excursions beyond the ore zone (both within the ore-bearing aquifer and in
overlying and underlying aquifers).
(4) Monitoring ground water chemical composition in wells surrounding and down gradient
of the extraction field.
(5) Monitoring the chemical composition of ground water up gradient of the extraction field
to determine if these waters are chemically stable over the course of the extraction effort.
• INJECTION WELL
o PRODUCTION WELL
A PRODUCTION ZONE MONITOR WELL
• OVERLYING AQUIFER MONITOR WELL
n UNDERLYING AQUIFER MONITOR WELL
(Source: NRC 2006)
Figure 4-9. Schematic Diagram of a Wellfield Showing Typical Injection/Production
Well Patterns, Monitoring Wells
Draft Technical Report
68
Revised Draft - November 26, 2012
-------
(Source: Card and Mahoney 2012)
Figure 4-10. Example of MODFLOW Predicted Potentiometric Surface during Active
Mining
+HHH+H-H-H
(Source: Card and Mahoney 2012)
Figure 4-11. Example of MODPATH Predicted Flow Paths During Active Mining
Draft Technical Report
69
Revised Draft - November 26, 2012
-------
4.5 Post-extraction Phase
The post-extraction monitoring system should be designed to assess the effectiveness of the
remediation process, assess when final remediation objectives have been met, and assure that the
affected aquifer is at steady state, long-term stability has been demonstrated, and the site is
ready for decommissioning. A system of wells located in the active treatment zone, as well as
outside the boundary of the affected area, is required to monitor the horizontal and vertical
ground water velocity and flow paths in the vicinity of the ISR site. The functions of a post-
mining monitoring system include:
(1) Measuring ground water chemical constituents to determine if and/or when the ground
water chemistry has returned to pre-ISR (baseline) compositions.
(2) Determining if additional chemical components have been added to the ground water as a
product of the extraction process (e.g., metals mobilized with the uranium).
(3) Demonstrating when the ground water chemistry has reached "stable" levels
(i.e., statistically equivalent compositions over an extended time period).
(4) Determining if post-mining restoration levels for ground water constituents have been
met.
4.6 Factors Affecting Post-mining Time Frames and Wellfield Stability
Post-restoration monitoring must be of sufficient duration to ensure that once ground water
chemistry appears to have reached acceptable restoration levels, these levels are at steady state
and the ground water system is at equilibrium. Steady-state restoration levels are not just for
uranium, but also for other hazardous constituents that may have been mobilized by ISR
operations, such as radium, manganese, and selenium. Chemical speciation and solubility, as
well as natural attenuation processes, must be understood in order to determine when the affected
aquifer has reached a steady-state condition. Both geochemical and advective-dispersive
modeling should be used as tools to assist in designing the most effective means of restoring the
aquifer. The approach that is generally taken to couple the geochemical processes with ground
water flow and contaminant transport is to first use MODFLOW to create a calibrated flow field
that reasonably resembles the actual field conditions. Once the flow conditions are established, a
transport code (e.g., MT3D) is applied to simulate simple geochemical behavior (e.g., sorption).
A more complex geochemical code is then applied (e.g., PHT3D) to simulate reactive transport
and surface complexation.
The environmental chemistry of uranium is largely dictated by its oxidation state; the solubility,
and therefore mobility, of uranium is greatest when it is in the U(VI) state. Because different
chemicals may be used during the restoration process than were used during ISR operation, the
chemical form of uranium or other hazardous constituents may differ during restoration. Since
most of the available computer codes do not have a method of calculating reaction rates, these
reactions may be unexpected, and the monitoring program must last long enough to
accommodate such unexpected conditions. There are a few sophisticated codes, however, that
allow the reactive transport processes to be coupled to the ground water flow regime, such as
PHAST or PHT3D. These codes have intensive data requirements, and EPA recognizes the
Draft Technical Report 70 Revised Draft - November 26, 2012
-------
importance of understanding the geochemical processes and has entered into a corporative
agreement with the U.S. Geological Survey under a Regional Applied Research Effort (RARE).
The main objective of this work is to provide a predictive model that describes the ground water
flow and geochemical changes along with longer-term transport of dissolved constituents during
and after the uranium ISR mining process (see also Section 4.3.7 below).
One of the most important factors affecting restoration times is that there may be significant
quantities of uraninite remaining after mining is complete (Figure 4-12). Post-mining sampling
and core analysis, as well as more complete tracking of the water balance and uranium mass
removal rates during mining, would assist in designing a more effective restoration.
Another factor affecting the post-monitoring time frame and wellfield stability is the form of
remediation used. Pump and treat techniques are common remediation approaches. One
consideration, however, are the large volumes of water required when pump and treat
technologies are used to restore the aquifer to pre-mining conditions. Issues related to both water
quality and quantity have resulted in EPA denying the Applicants Aquifer Exemption Request
until additional work is performed at the Goliad mine in Texas to demonstrate that nearby water
resources will not be adversely impacted for the next 75 years (EPA 2012).
Uraninite Remaining
(Source: Card and Mahoney 2012)
Figure 4-12. Example of PHT3D Predicted Post-mining Uraninite Concentrations
Geochemically based methods are promising as potential remedial alternatives, in that
significantly less water could be used during site restoration since strong reducing agents (e.g.,
hydrogen sulfide) are injected to induce chemical precipitation of the uranium from solution.
Monitored natural attenuation is another response action that may be effective in certain
situations. Natural attenuation processes include a variety of physical, chemical, and biological
processes that can act to reduce the mass, mobility, volume, or concentration of contaminants in
ground water. Attenuation mechanisms important at ISR sites are described in Table 3-4 and
include processes such as pH buffering and acid neutralization, adsorption at the mineral-water
interface, mineral precipitation, dilution, and biological activity.
Draft Technical Report
71
Revised Draft - November 26, 2012
-------
Pump and Treat
Alternative approaches included in pump and treat remediation are:
• Ground Water Transfer - This involves transferring ground water between the wellfield
starting restoration and another wellfield where uranium leach operations are beginning.
No liquid effluents are generated as water is transferred between one wellfield and
another.
• Ground Water Sweep - Injection of lixiviant is stopped, and the contaminated liquid is
pumped from the leaching zone via all of the injection and production wells. Fresh
ground water flows into the leaching zone from the outside, which displaces lixiviant in
the pore spaces. Typically, an ion-exchange system is used to process the sweep water,
which is disposed of either in evaporation ponds or via a deep injection well in
accordance with the site permit. The pumping rates are site specific, and the duration and
volume of water removed depend on the aquifer affected by the ISR. Due to
heterogeneities in the aquifers, ground water sweep alone may be insufficient and
uneconomical for complete restoration. In addition, ground water sweep may create oxic
conditions when up gradient waters enter the ore zone, making it more difficult to
re-establish chemically reducing conditions.
• Reverse Osmosis (RO) - To return ground water to baseline conditions, it is usually
necessary to remove contamination from the mined zone water while minimizing
disposal of waste liquids. RO, which involves passing the water being restored through
pressurized, semipermeable membranes, is a common way of treating ground water. The
RO treatment results in clean water or permeate that can be reinjected into the aquifer and
brine that is water with concentrated ions. The brine is usually sent to an evaporation
pond, injected into deep disposal wells, or dried (using an evaporator) for subsequent
disposal at a licensed facility.
• Permeate Injection - Many aquifers are characterized by porosity where ground water
with decreased mobility resides in regions of moderate to low permeability. It is very
difficult to remove all of the lixiviant and associated contamination from this portion of
the ground water, which will act as a source of contaminants, even after long periods of
pumping and treating.
Geochemically Based Techniques
Another component of aquifer restoration is accomplished by establishing a chemical
environment that alters the solubility of dissolved constituents. Chemicals may be added to
injection water in the latter stages of restoration to assist in re-establishing baseline conditions.
These methods typically invoke chemical reactions in which some species are reduced to a lower
valence state. Addition of reagents such as hydrogen sulfide and sodium hydrosulfide tend to
convert dissolved species such as uranium, selenium, molybdenum and arsenic to a lower
valence state with attendant reductions in solubility (NRC 2007, p. 17).
Draft Technical Report 72 Revised Draft - November 26, 2012
-------
Monitored Natural A ttenuation
Monitored natural attenuation (MNA) refers to the reliance on natural processes to achieve site-
specific remediation objectives within a reasonable time frame. These processes include
biodegradation, dispersion, dilution, sorption, and volatilization; radioactive decay; and chemical
or biological stabilization, transformation, or destruction of contaminants. The overall impact of
MNA at a given site can be assessed by evaluating the rate at which contaminant concentrations
are decreasing either spatially or temporally.
However, USGS has suggested a note of caution regarding the efficacy of natural attenuation
(NRC 2007, p. 17):
Because the ore zone typically is under chemically reducing conditions prior to
mining, it has frequently been argued or assumed that the natural reducing
conditions will return over a period of time. However, it is difficult to predict how
much time is required or even if the reducing conditions will be returned via
natural processes. The mining disturbance introduces a considerable amount of
oxidant to the mined region and may oxidize all the pyrite associated with the
original ore zone.
EPA has prepared a technical resource document (EPA 2007a and 2007b) that presents a four-
tiered assessment of MNA as a viable response action for selected metal, metalloid, and
radionuclide contaminants encountered in ground water. The assessment involves the following:
(1) demonstrating contaminant sequestration mechanisms; (2) estimating attenuation rates;
(3) estimating attenuation capacity of aquifer solids; and (4) evaluating potential reversibility
issues. EPA has a number of guidance documents that pertain to MNA (e.g., EPA 1999b) and
specifically to radionuclides (EPA 2010a).
Section 6.4 of this report presents additional details on MNA.
4.7 Modeling
4.7.1 Objectives and Conceptual Model Development
Modeling of ground water flow, contaminant fate and transport, and chemical speciation is often
utilized to predict the spatial and temporal behavior of the hydrogeochemical system. Ground
water flow and geochemical modeling is commonly implemented to assist in meeting different
objectives during each phase of the ISR process as described below:
Pre-Mining
• Establish background concentration of important ground water constituents and
determine geochemical constraints that may control contaminant concentrations, e.g.,
solubility constraints.
Draft Technical Report 73 Revised Draft - November 26, 2012
-------
Active Mining
• Determine spacing of injection and pumping wells.
• Optimize the monitoring well spacing to detect injection fluid excursions into non-mined
aquifer zone(s).
Aquifer Restoration
• Evaluate changes to hydraulic conductivities, gradients and flow directions.
• Estimate the number of pore volumes needed during site remediation activities to
adequately reduce contaminant concentrations.
• Estimate remediation times.
• Predict potential off-site impacts on water levels and quality.
Post-Remedial Monitoring
• Evaluate the long-term stability of the contaminants within the mined area, as well as any
that have migrated further down gradient.
• Establish a specific period of monitoring for ISR facilities once uranium extraction
operations are completed.
Prior to starting remediation of in-situ leach mining sites, modeling can be used to predict the
behavior of the ground water system during and after ground water restoration. To make such a
prediction, a conceptual model must be formulated that includes the most important physical and
geochemical processes that are occurring in the system at the end of restoration and that will
occur in the system in the future. In formulating such a model, three fundamental processes must
be included: ground water flow, solute transport, and chemical reactions. Secondary processes
such as microbial degradation may also be considered. Microbial action may not be important
during restoration since the ground waters are being moved through the system rather quickly,
but may have a more important role when the pumps stop and long-term stability is evaluated. In
addition, the initial conditions of certain physical and chemical variables in the system must be
specified, as well as any known changes to these variables that may occur in the future. It is
important to recognize that a model is only a tool for approximating a field system.
Data collected during pilot studies at an ISR field site should be useful in constructing the
conceptual model. For a ground water flow system, the nature of the conceptual model will
determine the dimensions of the physical model and the design of a grid for numerical
calculations. It is important to distinguish between the conceptual model of the hydrogeologic
system and a computer code. A computer code is a set of instructions for performing
calculations, whereas the conceptual model represents the physical and chemical understanding
of the system.
Draft Technical Report 74 Revised Draft - November 26, 2012
-------
4.7.2 Ground Water Flow and Contaminant Transport Modeling
After the conceptual model is developed, a computer code may be used to simulate the
controlling processes that affect ground water flow and chemical transport. Computer modeling
frequently used to meet the performance objectives at ISR facilities includes five primary types:
(1) ground water flow, (2) advective-dispersive transport, (3) advective transport (i.e., particle
tracking - no dispersion), (4) geochemical speciation, and (5) reactive-transport modeling.
These modeling approaches are described in greater detail below and in Section 4.7.3.
The ground water flow and advective transport modeling typically performed for ISR sites use
finite-difference techniques, which require that the ground water system be divided
("discretized") into finite-sized blocks or "cells." Each cell is assigned unique hydraulic
properties depending on the available field data and the goals for the analysis. In this way, the
model can accommodate complex features of the ground water system. The time represented by
the modeling effort must also be divided into discrete periods or "time steps." These steps must
be short enough to provide an accurate solution, but not so short that they require an excessive
number of calculations to run a simulation. The finite-difference method also requires that values
for hydraulic head be assigned at flow boundaries (referred to as "boundary conditions"), as well
as for the initial time period of the simulation (referred to as "initial conditions"). This is a
requirement for producing a unique solution with any numerical method that depends on
iteration, as does the finite-difference method.
The most popular computer code to simulate ground water flow is MODFLOW2000 (Harbaugh
et al. 2000). The output from this code is a three-dimensional flow field (an array of hydraulic
head elevations) representing average conditions in the model area.
Once the flow field is established, the most widely used contaminant transport code, MT3DMS
is used to predict future concentrations of selected constituents at ISR facilities (Zheng and
Wang 1999). MT3DMS does not explicitly simulate geochemical reactions but can be used to
simulate changes in concentrations of miscible contaminants in ground water, while considering
advection, dispersion, diffusion, and some aggregate chemical reactions (i.e., distribution
coefficient), with various types of boundary conditions and external sources and/or sinks. The
basic chemical reactions included in the model are equilibrium-controlled or rate-limited linear
or nonlinear sorption, and first-order irreversible or reversible kinetic reactions. Somewhat more
sophisticated, multispecies chemical reactions can be simulated by add-on reaction packages.
MODFLOW2000 and MT3DMS are commonly applied at ISR facilities to evaluate how the
average flow field, together with other transport parameters, affects chemical movement in
ground water and plume development from lixiviant sources. The chemical transport model is
often used to simulate the expansion of the plume, during both the time of active leaching
activities and the post-closure stage.
If diffusive and dispersive processes are not important to the transport analysis, the computer
code MODPATH is frequently used to compute three-dimensional advective flow paths. The
code uses output from steady-state or transient ground water flow simulations from MODFLOW.
MODPATH is described in USGS Open-File Report 94-164 (Pollock 1994). MODPATH uses a
Draft Technical Report 75 Revised Draft - November 26, 2012
-------
semi analytical particle tracking scheme that allows an analytical expression of the particle's flow
path to be obtained within each finite-difference grid cell. Particle paths are computed by
tracking particles from one cell to the next until the particle reaches a boundary, an internal sink
or source, or satisfies some other termination criterion.
After the modeling objectives are defined (Section 4.7.1), modeling of the ground water flow and
contaminant fate and transport generally follows the set of steps detailed below:
(1) Developing a conceptual model to guide creation of model attributes.
(2) Selecting an appropriate computer code(s) for the analysis.
(3) Establishing the time period represented by the model and the duration of subdivisions
of this period (time steps) required for modeling.
(4) Selecting a suitable model domain and determining the dimensional (horizontal and
vertical) limits of the analysis.
(5) Establishing the model structure by determining the number of model layers and the
grid spacing requirements for the flow analysis.
(6) Incorporating hydraulic boundaries and features, including the shape and characteristics
of constant-head boundaries, precipitation/recharge, and pumping/injection.
(7) Assigning hydraulic conductivity values.
(8) Specifying initial head values (ground water surface elevation).
(9) Evaluating and assigning appropriate model computational characteristics (e.g., solution
method, iteration limits, and convergence criteria) to enhance model stability,
computational efficiency, and solution accuracy.
(10) Evaluating the sensitivity of model results to changes in model parameters.
(11) Establishing the model structure, including determining the number of model layers and
the grid spacing requirements for the transport analysis.
(12) Assigning the characteristics of chemical sources (e.g., lixiviant), consisting of
dimensions, locations, concentrations, and time dependency.
(13) Assigning transport parameters, including the distribution coefficients, dispersivities,
and porosities.
(14) Defining chemical interactions among dissolved and solid phases.
(15) Developing remedial design scenarios and conducting chemical transport simulations
and exporting the observed concentrations at prespecified locations.
(16) Post-processing the data with Graphical User Interface (GUI) tools.
Although the ISR method of uranium mining has a less disruptive overall environmental impact
compared to open-pit mining, this mining method significantly alters the ground water chemistry
and flow patterns during mining. One of the common strategies at all of the ISR facilities is to
better understand the most probable fate and transport of uranium and other constituents during
Draft Technical Report 76 Revised Draft - November 26, 2012
-------
and after ISR operations. To achieve this goal, mathematical modeling of chemical reaction
kinetic equations or equilibrium thermodynamic equations can be used to describe chemical
interactions among dissolved chemical species, the dissolution of immobile solid phases, or the
formation and precipitation of new, immobile solid phases. The following section presents a
more detailed discussion of the role that geochemical modeling plays in the understanding of
important chemical processes at ISR facilities.
4.7.3 Geochemical Modeling
Geochemical reactions along ground water flow paths can lead to regional variations in water
composition that evolve in the direction of flow. Iso-concentration contours of reacting dissolved
constituents drawn on maps of water composition tend to align perpendicular to the direction of
ground water flow. It is these geochemical reactions that have created the sandstone-hosted
uranium roll-front deposits along the interface between oxidized and reduced sandstones. The
geochemistry associated with these deposits is complex and variable. The deposits contain a
mixture of U+6 minerals on the oxidized side of the front and reduced U+4 minerals on the
reduced side of the front (Harshman 1974). Oxidizing ground water flowing through sandstones
transports uranium U+6 in solution down dip until reductants in the host sandstones precipitate
uranium as U+4 minerals. Associated elements are often found distributed across the roll in zones
determined by their redox potential and solubility in alkaline, oxidizing ground water that comes
into contact with pH neutral, reducing sediments at the reaction front (Deutsch et al. 1985;
Harshman 1972). Ground water within the ore zone of roll-front deposits often contains high
concentrations of uranium and its daughter products, as well as elevated selenium, arsenic, lead,
and other dissolved constituents (Johnson et al. 2010).
Modeling of any type does not lead to a unique solution, but the number of possibilities are
reduced with greater amounts of carefully collected field data. Martin et al. (2005) summarized
the benefits and limitations of geochemical modeling as follows:
Benefits
Provide insight into potential future conditions.
Determine which variables are most important in determining future
conditions.
Assess potential effects of uncertain parameters.
Establish objectives and test conditions for field and laboratory studies.
Integrate available information.
Limitations
Insufficient input data.
Modeling can be challenging and results misinterpreted.
Uncertain and variability of the results.
Difference between modeled and actual field conditions.
Draft Technical Report 77 Revised Draft - November 26, 2012
-------
Approaches to Geochemical Modeling
All geochemical models are based on principles of mass conservation (mass balance accounting).
Mass is neither created nor destroyed in the system, but transferred between solid, aqueous, and
gaseous phases. Geochemical models can be generally sorted into two distinct categories,
however, depending on the extent to which they incorporate transport processes. Models that do
not consider transport processes are referred to as "geochemical reaction models" or simply
"batch models." Models that consider both transport processes and geochemical reactions are
referred to as "coupled transport and reaction models." Three basic modeling approaches have
been used with geochemical data: inverse geochemical modeling, forward geochemical
modeling, and reaction path modeling.
Inverse modeling attempts to establish reaction mechanisms that explain measured chemical
changes that occur as water composition evolves along a flow path; it assumes that a water flow
path is known and that water samples have been analyzed along that flow path. Such data can
then be converted into amounts of minerals dissolved or precipitated along that flow path.
Several assumptions are still made regarding the choice of minerals and their relative proportions
contributing to the water chemistry, but the calculations are constrained with actual data. Inverse
modeling can also be done without any recourse to kinetic or thermodynamic data, in which case
it represents a relatively simple mass balance calculation. When speciation and thermodynamic
and kinetic properties are included for additional constraints, the possible reactions become quite
limited, and the modeling is much more meaningful. Inverse modeling calculations provide a
way to evaluate the most probable chemical reactions between water and minerals.
Forward modeling is also known as "simulating" (i.e., potential reactions between rock and
water are simulated from initial conditions of a known rock type and composition). Reactions are
allowed to proceed in equilibrium or kinetic or combined modes. Changes in temperature and
pressure can be invoked, changes in water flow rate can be assessed, and minerals can be
allowed to precipitate as they reach equilibrium solubility or dissolve as they become
under saturated. Potential reactions can be simulated to see what the consequences are. This type
of modeling is the least constrained. A great many assumptions are either invoked as input data
or invoked as dictated by the program that may not apply to the specific system being simulated.
This approach assumes the modeler has a significant amount of information on the ability of
minerals to maintain equilibrium solubility or their rates of reaction. A typical example of
forward modeling would be the calculation of the final water composition in an aquifer where
infiltrating rainwater is allowed to equilibrate with calcite and dolomite (as might occur in a
limestone aquifer).
Reaction path modeling or mass transfer modeling is dynamic in the sense that it allows the
simulation of how changes in water and mineral phase composition occur over time as defined
primary minerals are dissolved incrementally. At each step in the calculation, the aqueous
speciation is calculated, and secondary minerals are dissolved or precipitated in order to maintain
equilibrium. These models have been widely used to evaluate the chemical weathering processes
that occur in natural systems (diagenetic processes). The gradual weathering of igneous rocks to
produce clay minerals is a good example of a process where reaction path modeling may be
useful. As these models consider the dissolution of primary minerals as a stepwise process, the
Draft Technical Report 78 Revised Draft - November 26, 2012
-------
variable of time is not included implicitly in the calculations. If kinetic data are available,
however, that can be used to relate reaction progress to time, and the aqueous composition may
be calculated as a function of time in a kinetic geochemical reaction model.
Mathematical Formulation
In all geochemical models, the reactions that describe the aqueous composition must be defined
in terms of a minimum set of fundamental basis species that are required to describe all the free
and derived species (complexes) present in the aqueous solution (e.g., f^O, H+,CO3~ ,OH"). The
basis species do not need to be real species that exist in the solution; the only limitations are that
they are mutually independent (i.e., they cannot be described in terms of combinations of each
other) and that they provide a complete stoichiometric description of the system.
"Speciation" refers to the distribution of chemical species or elements among the different
possible forms or species. Aqueous speciation is the distribution of chemical species among
dissolved free ions, ion pairs and triplets, and other complexes. This concept is important
because research has shown that some processes, including mineral precipitation and dissolution,
biological uptake and toxicology, and sorption, are all affected by speciation. Some species, such
as redox species, must be determined analytically. This is because most geochemical modeling
codes erroneously assume that redox equilibrium is maintained, while in reality, disequilibrium
among redox species is the rule, not the exception.
Most geochemical reaction programs are based on an approach in which the conservation of total
component concentrations is combined with a description of chemical equilibrium. Chemical
equilibrium may be computed in terms of Gibbs' free energy minimization or in terms of mass
action equations involving equilibrium constants. The method of Gibbs' free energy
minimization is generally regarded as being more mathematically robust than the method using
equilibrium constants. Because of the lack of reliable and internally consistent Gibbs' free
energy data, however, geochemists have tended to favor the equilibrium constant method. The
vast majority of programs available today are therefore based on the equilibrium constant
approach.
Aqueous speciation results are used for a variety of modeling objectives, including modeling of
saturation-index calculations for mass transfer, modeling of mineral precipitation and
dissolution, modeling of adsorption and desorption, and reactive-transport modeling.
For speciation reactions, the computer code solves a reduced set of simultaneous nonlinear
equations that define equilibrium for a water, solute, gas, mineral, ion-exchanger, and surface-
complexer chemical system. Equilibrium is based on an ion-association model for the aqueous
phase and mass-action equations for mineral, gases, exchangers, and surface complexers. The
complete set of equations includes a mole balance equation for each element in the system;
mass-action equations for each aqueous species, each gas component, each mineral, each
exchange species, and each surface complexer; an activity coefficient equation for each aqueous
species; a charge-balance equation for the aqueous phase; a charge-balance or charge-potential
equation for each surface complex; an equation for the activity of water; and an equation for the
ionic strength of a solution. Subsets of this set of equations are solved for a particular
Draft Technical Report 79 Revised Draft - November 26, 2012
-------
geochemical calculation. The equations are solved by a modified Newton-Raphson calculation.
The modification involves the use of an optimization routine based on linear programming.
During the iterative Newton-Raphson process, some of the equations are included as objective
functions rather than strict equalities. This approach is useful for determining the stable set of
minerals and the presence or absence of a gas phase in a chemical system; it also makes the
numerical algorithm more robust. The solution to the equations provides the activities and
molalities of each aqueous species and the moles of each mineral, gas component, exchange
species, and surface species present in the system.
In inverse modeling, one aqueous solution is assumed to react with minerals and gases to
produce the observed composition of a second aqueous solution. The inverse model calculates
the amounts of these gases and minerals from the difference in elemental concentrations between
the two aqueous solutions. It is also possible to determine mixing fractions for two or more
aqueous solutions and the mole transfers of minerals necessary to produce the composition of
another aqueous solution. Inverse modeling is based strictly on a mole-balance approach and
does not rely on the ion-association model except to determine the total number of moles of each
element and redox state in each aqueous solution. The inverse model is formulated including
uncertainty in each analytical datum. A linear set of equations is formulated including mole
balance for each element and element redox state in the system, a charge-balance equation for
each aqueous solution, and a water-balance equation. In addition, inequality constraints are
included to ensure that any adjustments to the analytical data are smaller than the uncertainties
and to constrain the sign of mole transfers of mineral (if specified). The system of equalities and
inequalities is solved by an optimization routine based on the Simplex method. An additional
algorithm is used to find all sets of minerals that are feasible solutions to the inverse problems.
Processes Simulated
Modeling of mineral precipitation and dissolution and gas-transfer reactions can take place
conceptually in one of three possible systems: equilibrium state, steady-state, or transient state.
The equilibrium state assumes that the system under investigation is isolated from any external
exchanges of energy or mass. Although an unrealistic concept, equilibrium state is actually quite
practical because many reactions approximate equilibrium even though there are gradients in
water pressure or temperature. For example, in many ground waters, calcite and gypsum quickly
reach their equilibrium solubility. Even with gradients in CC>2 pressure or mixing with other
sources of sulfate, these minerals adjust to maintain saturation, and the assumption of
equilibrium may be valid. In addition, even when geochemical reactions of interest do not reach
equilibrium rapidly, such reactions may achieve equilibrium over the time scale of the modeling
simulation (i.e., the life of a mine and beyond). Therefore, the majority of geochemical modeling
can be conducted under the assumption of equilibrium conditions.
The most simplistic geochemical models are empirical sorption models and describe
experimental adsorption data without any theoretical basis. These models rely on adsorption
isotherms that plot the concentration adsorbed to the solid surface versus the concentration in
aqueous solution for different total concentrations of a chemical species. One of the most widely
used adsorption isotherm equations is a linear function written in terms of the distribution
coeffient, Kd:
Draft Technical Report 80 Revised Draft - November 26, 2012
-------
x =
where x is the amount of chemical species adsorbed per unit mass of solid and c is the
equilibrium solution concentration of the chemical species.
Due to the complexity of the processes involved, a single partition or distribution coefficient is
often used that describes the degree to which the contaminant's transport is retarded relative to
water. This approach starts with defining the retardation factor:
-
f~~c
where:
Rf = the retardation factor
vp= the velocity of water through a control volume
vc= the velocity of contaminant through a control volume.
Langmuir (1997) noted that the retardation factor is related to the distribution coefficient
through the expression below:
R =\+HLK
"„
where:
pbis the porous media bulk density, and
we is the effective porosity at saturation given as a volume fraction.
In addition to empirical sorption models, chemical surface complexation models have been
developed to describe the potentiometric titration and metal (e.g., uranium, arsenic, selenium,
molybdenum, iron, and vanadium) adsorption data at the oxide-mineral solution interface.
Surface complexation models of the solid-solution interface share at least six common
assumptions: (1) surfaces can be described as planes of constant electrical potential with a
specific surface site density; (2) equations can be written to describe reactions between solution
species and the surface sites; (3) the reactants and products in these equations are at local
equilibrium and their relative concentrations can be described using mass law equations;
(4) variable charge at the mineral surface is a direct result of chemical reactions at the surface;
(5) the effect of surface charge on measured equilibrium constants can be calculated; and (6) the
intrinsic (i.e., charge and potential independent) equilibrium constants can then be extracted from
experimental measurements (Koretsky 2000).
More sophisticated geochemical modeling approaches allow for the precipitation and dissolution
of gases and minerals, as well as the possibility of fixing the activity of specified components
(the hydrogen ion activity, pH, for example). Reaction types that can be handled usually include
complexation, ion-exchange, redox reaction, precipitation/dissolution, surface complexation, and
other kinds of adsorption. The major limitation is the quality and availability of thermodynamic
data for carrying out reaction calculations. Many programs contain databases of relevant
aqueous, gaseous, and mineral phase reactions, and the more sophisticated programs can
Draft Technical Report 81 Revised Draft - November 26, 2012
-------
automatically select mineral or gaseous phases that are likely to precipitate and include them in
the calculations. Some programs can be used to simulate titrations, evaporative processes, and
mixing of different solutions or to perform isotope mass balances. Mass balances based on
radiogenic isotopes are used primarily for estimating the age of ground water (i.e., the time
elapsed since it entered a ground water system). Mass balances that consider stable isotopes are
used to understand the source of water or the processes that may have influenced the chemical
properties of the water over time.
Two general approaches are used in geochemical reaction models to calculate activity
coefficients of aqueous species. The first type consists of the Debye-Hiickel equation and its
variant, the Davies equation, and its extended B-dot equation form. This approach limits the field
of applicability for these models to solution ionic strengths less than or equal to that roughly
corresponding to seawater (Parkhurst 1995). The second approach involves the use of Pitzer
equations, which can be applied accurately to systems of high ionic strengths such as brines and
highly saline ground waters (Pitzer 1979). Due to the high TDS concentrations within the mined
area at the ISR sites, the application of geochemical models that include the Pitzer equations is
important to accurately model geochemical reactions. However, because redox reactions,
especially for the Fe(II)/Fe(III) system and the aluminum system, have not been fully
parameterized for the Pitzer model, use of the Pitzer method currently will not adequately
address geochemical reactions involving iron or aluminum associated with certain geochemical
systems.
An increasing number of programs allow the simulation of kinetically mediated processes. These
programs generally require user input to define kinetic parameters and sometimes the kinetic
reaction equations themselves. As in the Pitzer method for calculating aqueous phase activities, a
noted problem is the lack of kinetic data in the literature for many important mineral reaction
processes.
Reactive Transport (CoupledModels)
Coupled transport and reaction models differ from the geochemical reaction models described
previously in that transport processes are included explicitly in the mathematical formulation of
the model. The idea is to couple flow models with chemical reaction models to determine the
effects of flow on reactions, including the effects of dispersion.
Coupled transport and reaction models can be used to simulate how a geochemical system
evolves over time along a fluid flow path in one, two, or even three dimensions. Like
geochemical reaction models, coupled transport and reaction models are based on the principle
of mass conservation. Whereas the mathematical formulation of a geochemical reaction model
generally regards a single control volume that is formally decoupled from flow considerations,
coupled transport and reaction models discretize the flow medium into a network of
interconnected control volumes. Reactive-transport modeling for ground water has also
progressed substantially over the last three decades, and many of the recent codes have been
applied to ISR mine sites. Mayer et al. (2003) provide an overview of the theoretical foundations
for ground water reactive-transport modeling, methods of coupling flow with reaction, and the
various codes available.
Draft Technical Report 82 Revised Draft - November 26, 2012
-------
Critical to the success of a coupled transport and reaction model is detailed knowledge of the
hydrology of the site to be modeled. Frequently, it is not possible to obtain the necessary amount
or quality of data to satisfactorily characterize the subsurface system for the purpose of a reliable
predictive simulation. This problem arises largely from issues of heterogeneity. Heterogeneity in
subsurface soils and rock manifests itself in the form of preferential flow paths, fracture zones,
regions of variable hydraulic conductivity and porosity (layered sedimentary rocks and soils), as
well as stagnant zones (clay lenses and other flow-isolated porosities in the rock matrix. Other
factors that may affect the reliability of coupled transport and reaction models are the transient
nature of contaminant sources and the variable boundary conditions relating to the ISR injection
and pumping wells.
Some programs have been developed that can simulate heterogeneous reaction systems. These
models consider alterations that may occur in the distribution of minerals in the system under the
influence of reactive-transport processes. The mathematical formulation of models for
heterogeneous reaction systems is much more complicated than that for homogeneous reaction
systems, as zones of dissolution and precipitation form and slowly advance. One of the problems
associated with the simulation of heterogeneous reaction systems is the necessity to track the
position of these mineral reaction fronts over time. The programs that simulate heterogeneous
reaction systems are frequently unable to simulate the entire suite of geochemical reactions that
can be simulated by nontransport-enabled geochemical reaction programs.
In general, it is difficult to accurately simulate kinetic processes involving heterogeneous
reactions. Kinetic interactions with solid phase materials are usually quite strongly dependent on
the mineral surface area exposed to pore water ground water conditions (i.e., complexing ion
concentrations, pH and redox potential), and the residence time of water in the random pores and
fractures that characterize most geological media. The exposed mineral surface area and the
porosity of the medium change during diagenesis as a result of the precipitation and dissolution
of various minerals. The exposed surface area of some minerals may decrease as a result of the
precipitation of other minerals that block their access to the pore water.
Local changes in the porosity of the medium may also give rise to preferential flow paths.
Because of relationships between mineral surface area and porosity, the creation of preferential
flow paths can lead to the formation of fingered mineral alteration zones. These processes are
virtually impossible to predict. Fortunately, however, it is rarely necessary to know specific
details about the formation of fingered zones, and it is often sufficient to assume a relatively
homogeneous porous medium. Although small-scale heterogeneities are often neglected, some
information about mineral surface area is still required in order to estimate mineral reaction rates.
Mineral dissolution and precipitation rates are frequently modeled using semiempirical
approaches such as the transition state theory (Lasaga 1981; Aagaard and Helgesson 1982).
Conceptual models for reactive-transport modeling are necessarily complex and will likely
require that alternative conceptual models be considered in order to examine the range of
simulation results and the sensitivity of predictions to conceptual model error. Conceptual
models for reactive-transport modeling represent the scientific understanding of processes
controlling the movement and transformation of system components, including contaminants, for
a specific water rock system (Davis et al. 2004). For example, a conceptual model for the ISR
mined region might include knowledge of (1) initial spatial distribution of chemical species
Draft Technical Report 83 Revised Draft - November 26, 2012
-------
(including uranium, arsenic, iron, sulfur, selenium) and mineralogy, (2) hydrologic sources and
sinks, porosity, and spatial dependence of hydraulic conductivity, and (3) aqueous solute
speciation and chemical reactions controlling phase distribution. Alternative conceptual models
for ground water restoration at ISR facilities might include different initial concentrations of
various minerals or variable redox status of ground water flowing into the subsurface region that
was mined.
Predictive Modeling Strategies
USGS, in conjunction with EPA, is in the process of developing modeling strategies to address
three primary questions (Johnson et al. 2010): (1) How well do identified aquitards limit ground
water flow between aquifers? (2) What is the ground water quality at the end of mining after
restoration efforts are complete? and (3) What are the long-term fate and transport of any ground
water contaminants away from the mined zone? These strategies will be generic for any uranium
ISR mine, but will be applied to a current site in South Dakota that is in the permitting phase.
This research is designed to assist EPA, mining companies, the general public, and other
stakeholders with specific strategies for understanding and modeling subsurface hydrogeology
and geochemistry. The types of information that are important to support detailed modeling
studies include the development of geologic models based upon exploratory drill holes, solid
phase mineralogy acquired from core analysis and ground water chemistry obtained from ground
water sampling.
Once the site data are successfully collected, the researchers outline four general components of
a strategy beginning with the development of a conceptual model that establishes the basic
hydrogeologic and geochemical system, using available data and professional expertise. The
conceptual model includes information such as ground water flow direction and velocities,
boundary conditions, and current ground water and solid-phase geochemistry. The second step
involves numerical modeling with reactive-transport models, which simulate ground water flow
and geochemical interactions between the aqueous and solid phases. For this step to be
successful, initial solid phase geochemistry is required. The third step applies modeling to better
understand the potential impact of the mining activities on surrounding ground water quality
under various design options. The final step in the process focuses on the evaluation of the model
limitations and uncertainty in the model input parameters (e.g., geochemistry and hydraulic
conductivities).
Examples of Major Codes
The computer codes that are applied to modeling at ISR facilities are typically selected to be
consistent with the modeling objectives and the available data. For instance, if the intention is to
calculate uranium migration velocities and travel times, then it may be appropriate to apply a
code that utilizes an empirical sorption model (Section 4.7.3), rather than a geochemical reaction
model based upon thermodynamics. Estimating restoration time frames, however, may require a
geochemical reaction path or mass transport code(s). The discussion below provides a general
overview of the types of computer codes that are available to explicitly simulate
thermodynamically based reactions. Although the complexity of the codes varies, their
applicability will be largely dependent upon the modeling goals and whether the underlying
kinetics and thermodynamics are well understood.
Draft Technical Report 84 Revised Draft - November 26, 2012
-------
Inverse Modeling. Inverse modeling is used to explain the observed chemical and isotopic
evolution of natural (or contaminated) waters, rather than to predict future compositions. This
modeling approach applies geochemical mass balances to the observed chemical and isotopic
composition of evolutionary ground waters to estimate masses of mineral and gas transfer in
water-rock systems. Inverse geochemical modeling software includes BALANCE (Parkhurst
et al. 1982), NETPATH (Plummer et al. 1994), PHREEQC (Parkhurst 1995; Parkhurst and
Appelo 1999), PHREEQCI (Charlton et al. 1997; Charlton and Parkhurst 2002), SPREADBAL
(Bowser and Jones 2002), Geochemist's Workbench® (Bethke and Yeakel 2009), and CrunchFlow
(Steefel 2009).
The inverse modeling capabilities of the PHREEQC and PHREEQCI codes consider the
uncertainties associated with the definition of initial and final solution compositions (chemical
and isotopic) and with the isotopic composition of reacting phases. The NETPATH code offers a
complete suite of adjustment models for C-14 dating.
Forward Modeling. The "forward modeling" approach has been extended to geochemical
transport codes capable of simulating ground water flow and the advection and dispersion of
solutes, coupled with a complex array of geochemical processes. Completely specified isotopic
reactions have been incorporated into geochemical mass-transfer and mass-transport codes
[specifically into PHREEQCI and PHAST (Thorstenson and Parkhurst 2002, 2004)], allowing a
forward modeling description of the isotopic evolution of a ground water system, along with its
concurrent chemical evolution.
Progress has also been made to numerically simulate the coupling of ground water flow, solute
transport, and geochemical processes. Geochemical mass-transport codes [e.g., MST1D
(Engesgaard and Kipp 1992); PHREEQC/PHREEQCI; PHAST (Parkhurst et al. 2004)]
incorporate all the limitations and uncertainties associated with the use of (1) geochemical
reaction codes and (2) nonreactive solute transport codes.
Reaction Transport Codes. Glynn (2003) contrasts the use of highly simplified reactive-
transport codes with the PHREEQC geochemical transport code, a code limited to a
one-dimensional description of flow and transport, but with a more complex, conceptually more
accurate description of sorption mechanisms and multispecies geochemical processes. Inverse
and forward geochemical modeling were conducted, including a three-dimensional geochemical
transport model using the USGS code PHAST.
Table 4-1 shows some of the more popular codes used primarily for ground water geochemistry
but also for sites affected by mining. More detail on geochemical modeling, modeling codes, and
associated uses and limitations is presented in Alpers and Nordstrom (1999), Mayer et al. (2003),
and Maest and Kuipers (2005).
Draft Technical Report 85 Revised Draft - November 26, 2012
-------
Table 4-1. Summary of Commonly Applied Geochemical Modeling Codes
Codes
CrunchFlow
EQ3/6
Geochemist's Workbench
MIN3P
MINTEQA2
NETPATH
PHAST
PHREEQC
PHRQPITZ
SOILCHEM
SOLMINEQ.GW
WATEQ4F
Type
Lawrence Berkeley National Laboratory supported code:
Multicomponent Reactive Flow and Transport.
Lawrence Livermore National Laboratory code: mass
transfer and reactive transport
University of Illinois code: mass transfer, saturated flow
Waterloo code: saturated and unsaturated flow
EPA supported code: speciation and mass transfer
USGS codes: mass transfer and reactive transport
USGS codes: mass transfer and reactive transport
USGS codes: mass transfer and reactive transport
USGS codes: mass transfer and reactive transport in high
ionic strength solutions
University of California: Chemical Speciation
USGS code: mass transfer and high temperature
USGS code: speciation and low temperature only
Reference
Steefel 2009
Wolery 1992
Bethke 2002, 1996, Bethke
and Yeakel 2009
Mayer et al. 2002
Allison etal. 1991
Plummeretal. 1994
Parkhurst et al. 2004, 2010
Parkhurst and Appelo 1999
Plummer and Parkhurst 1990
Sposito and Coves 1988
Perkins etal. 1990
Ball and Nordstrom 1991
Modeling Case Histories
Ground water modeling is often performed at ISR facilities to gain a better understanding of
various processes such as mineral dissolution/precipitation, oxidation/reduction and
adsorption/desorption. The material presented below documents historical modeling activities
together with research currently being conducted. The material is excerpted and paraphrased
from the original documents.
Modeling of the Moore Ranch ISR Project (NRC 2010)
The simulated sand unit was assumed to be partially saturated over the proposed license area. To
assess potential drawdown, a ground water flow model was developed to simulate partially
saturated conditions. The model was created within the Ground Water Vistas GUI platform and
applied MODFLOW-SURFACT (Version 3.0). The model was calibrated to site-specific
conditions and verified by site-specific field pumping test data. The model analyzed drawdowns
during various phases of ISR production and aquifer restoration. The model was also used to
estimate the potential impact of the simulated drawdown on private well users within 2 miles of
the facility boundary. The estimated drawdown was determined to have a negligible impact on
private well yield. The model was also used to determine the impact of production on water
levels in other water-bearing sandstone lenses. This modeling effort was only focused on the
physical ground water flow processes and did not consider geochemical reactions.
Consideration of Geochemical Issues in Ground Water Restoration at Uranium ISR Mining
Facilities (NRC 2007)
This report discusses various ISR topics including developing and applying a conceptual model
that considers the ground water flow, solute transport, and geochemical reactions associated with
the Ruth ISR. The modeling was designed to provide a quantitative and dynamic method for
estimating the number of pore volumes associated with ground water restoration as a function of
Draft Technical Report
86
Revised Draft - November 26, 2012
-------
both historical conditions and potential variations (i.e., under different assumptions of future site
conditions). Once the conceptual model was developed, the data collected from the site were
input into PHREEQC to estimate the number of pore volumes that must be removed to return the
system to initial conditions.
A series of reactive-transport simulations using ground water restoration data from the Ruth ISR
pilot-scale study was used to evaluate variations in the geochemical processes. The calculations
showed that a computer code like PHREEQC can be used to make predictive calculations of how
different geochemical conditions may affect evolving water quality during geochemical
restoration.
Irigaray Mine Wellfield Restoration Report, Johnson County, Wyoming (COGEMA 2005)
A ground water flow and transport model was developed to assess fate and transport of residual
constituents derived from the Irigaray ISR mine. The objective was to demonstrate that residual
concentrations would be below regulatory standards at prescribed observation points. The model
was used to evaluate continued migration under steady-state conditions, without pumping or
injection. Therefore, the emphasis was on adequately representing steady-state advective-
dispersive flow within and across the production zone. The approach taken was to develop a
model that predominantly depended on advective-dispersive transport of constituents and
minimized reliance on geochemical reactions. The parameters required included hydraulic
conductivity, hydraulic gradient, dispersivity, and effective porosity. These parameters were
quantified for the site and were incorporated into the model. Dispersivity values were based on
the scale of the site/plume and literature review.
A distribution coefficient (Kd) was included for some constituents using the lowest reasonable Kd
available from the literature. Uncertainty was addressed in model parameters with a sensitivity
analysis. The codes used to develop the model included:
• MODFLOW: for simulation of the flow field.
• MODPATH: for simulation of ground water flow paths.
• MT3DMS: for simulation of transport of site-derived constituents.
Although MT3DMS was calibrated to simulate migration of selenium, manganese, uranium,
Ra-226, and TDS, it uses the simplified distribution coefficient approach to calculate retardation
factors as described in Section 4.7.3. The modeling results demonstrated that remaining
concentrations at distances 400 feet down gradient of the wellfield were below the regulatory
standards for all the constituents.
Ross ISR Project, NRC License Application, Crook County, Wyoming (Strata Energy 2010,
Addendum 2.7H Ground Water Model)
The model was developed to analyze the potential direct, indirect, and cumulative hydrological
effects of the project on both regional and individual wellfields. As stated in the document, the
primary goals of the regional ground water model were to (1) identify potential impacts (if any)
to adjacent water rights, (2) estimate long-term impacts from ISR operation, and (3) identify
Draft Technical Report 87 Revised Draft - November 26, 2012
-------
potential impacts to the surficial aquifer and surface impoundments. Modeling goals for
individual wellfields were to:
• Estimate adequate perimeter monitoring well offset/setback distances for the wellfield.
• Demonstrate the ability to identify and remedy a lateral excursion (i.e., lixiviants moving
past the monitor wells).
• Optimize the wellfield design and pumping rates including bleed rate.
• Evaluate restoration time/efficiency.
The hydrogeology of the area ground water use was presented as a basis for the model. The
numerical ground water model was the USGS modular finite-difference model MODFLOW and
Ground Water Vistas was used as the pre- and post-processor. These codes were selected
because of their wide use and acceptance by the industry and regulators. Hydraulic parameters
used in the model were based on site data or taken from the literature. The model was first
calibrated to a steady-state solution based on pre-1980 conditions. Once a satisfactory calibration
to steady-state conditions was achieved, a transient calibration was conducted with the goal of
matching the drawdown that had occurred over 30 years due to withdrawals from industrial
wells. To assist in the calibration, PEST (a model independent parameter estimation program)
was used.
The calibrated model was used to simulate ISR operations within the Ross Project area. The ISR
simulation was a generalized scenario based on current mapped mineralization. Impacts were
also determined along with recovery simulation and flare evaluation. The general conclusion
from the modeling was that impacts of the facility would be minor. Lastly, the model is expected
to be a useful tool for the final wellfield planning and operations and assist in balancing
wellfields, progression planning, and bleed rate optimization.
Wellfield Restoration Report, Christensen Ranch Project Wyoming (COGEMA 2008a)
While this report presents a transport assessment:
... no groundwater modeling specific to the Christensen Ranch MUs is included in
this report. Christensen Ranch site conditions, including the constituents of
concern, are similar to those at the Irigaray Mine. Groundwater modeling
included in the Irigaray Mine Aquifer Restoration Report (COGEMA 2003) is
referenced where applicable to the Christensen Ranch Site.
The transport assessment comprised the geochemical assessment and hydrological assessment.
The geochemical component of the transport assessment addressed the physical and chemical
behavior of constituents of concern under the prevailing environmental conditions at the site.
These included uranium, radium, iron, manganese, selenium, and sulfate. In addition to the
geochemical assessment, the effects of long-term ground water flow, including advective mixing
and dispersion, on constituent concentrations were considered under the transport assessment.
Direction and velocity of ground water flow are critical hydrologic factors with respect to solute
Draft Technical Report 88 Revised Draft - November 26, 2012
-------
transport. Determination of the direction of ground water flow was based on water-level data that
were routinely collected from the monitor well networks.
As previously stated, no ground water modeling was performed specifically for the Christensen
Ranch site. However, extensive modeling of ground water flow and solute transport was
performed for the Irigaray Mine site located 6 to 10 miles northwest of Christensen Ranch. The
Irigaray Aquifer Restoration Report includes those modeling results, which were accepted by the
Wyoming Department of Environmental Quality (WDEQ) and NRC. These results apply in a
general sense to Christensen Ranch.
The model focused on the impacts of advective mixing on constituent concentrations migrating
from the Irigaray site, but did not address the effects of geochemical processes along the flow
path. The results of the model indicate that the maximum concentration at a distance of 400 feet
from the wellfield was approximately 7 times lower than the initial average wellfield
concentration, after subtracting the average background concentration. The background
concentration was subtracted from the affected concentrations to normalize the data. The monitor
well ring is located 400 feet from the wellfield.
Highland Uranium Project A-Wellfield
The operator used modeling to show that natural attenuation processes would reduce ground
water contamination to acceptable levels. (See Attachment B for details.) The attenuation
modeling involved both ground water flow modeling with MODFLOW and PHREEQC
geochemical modeling. The calculations showed that a maximum of 15 years would be required
to achieve the full benefits of natural attenuation. To support the modeling results, the operator
was required to perform semiannual monitoring of four wells (a "hot spot" well with elevated
levels of uranium and selenium, an up gradient well, a down gradient well, and a lateral well)
beginning in 2004. At the current time, field measurements indicate that the uranium and
selenium concentrations are stable, but not declining, as would be expected from natural
attenuation and predicted by the geochemical modeling.
International Mine Water Association
A review of the meetings of the International Mine Water Association (IMWA) uncovered two
relevant publications discussing predictive modeling for ISR facilities (Johnson et al. 2010;
Johnson 2011). As observed by Dr. Johnson, this effort is still preliminary. Below is a review of
the two papers.
Predictive Modeling Strategies for Operations and Closure at Uranium In-Situ Recovery Mines
(Johnson et al. 2010) was presented at the IMWA 2010 symposium, "Mine Water and Innovative
Thinking." In this paper, the authors present a predictive strategy, which will be applied at the
IRS facility in Edgemont, South Dakota (the proposed Dewy-Burdock facility in Fall River and
Custer Counties).
The following describes the steps proposed by Johnson et al. 2010:
Draft Technical Report 89 Revised Draft - November 26, 2012
-------
First, a conceptual model must be established to understand the basic
hydrogeologic and geochemical system, based on available data and professional
expertise. Such a conceptual model includes information such as groundwater
flow direction, boundary conditions, along with current groundwater and solid-
phase geochemistry.
Second, predictive modeling using numerical reactive transport models can be
used to simulate future groundwater conditions (during mining, restoration, and
post-restoration). This requires the coupled simulation of groundwater flow and
geochemical reactions using such models as PHT3d (Prommer 2002), which
simulates groundwater flow using MODFLOW (Harbaugh and others 2000) and
geochemistry using PHREEQC (Parkhurst andAppelo 1999). In any reactive
transport modeling, input data linking the solid-phase mineralogy to the
groundwater quality is very important to understand the rock-water interaction.
For predictive modeling purposes, the collection of solid-phase geochemistry
before mining is required. For final model calibration, postmining solid-phase
geochemistry is optimal. Theoretical reactive transport simulations of uranium
ISR mining have been examined (Davis and Curtis 2007 [NRC 2007]); however,
field applications have been limited.
Third, predictive modeling can be used to evaluate the impact on surrounding
groundwater quality under the proposed mine plan design and to evaluate
possible design alternatives. Since many uranium ISR-amenable deposits occur
within sandstones that are drinking water aquifers outside of the ore zone,
protection of groundwater quality is of great importance. Predictive reactive
transport modeling provides a tool for evaluating potential impacts on
surrounding groundwater quality based on initial mine plans. This is part of the
second modeling strategy discussed above, but more importantly, alternate well
field design and possible restoration procedures can be evaluated before
finalizing any mine operation and closure plans.
Fourth, model limitations should be evaluated to provide a reasonable range of
prediction uncertainties. This step involves the evaluation of uncertainties in the
model input parameters, such as geologic layering (Johnson andFriedel 2009)
and water chemistry. For example, the integrity of the confining zone should be
evaluated based on any uncertainty in geologic logs and can be tested using
multiple geologic conceptual models. In addition, any predictions of long-term
contaminant transport should provide adequate prediction uncertainties based on
the uncertainties of the input data. Multiple conceptual models provide a range of
potential groundwater quality impacts. This provides valuable feedback for the
collection of additional data, which can assist in reducing uncertainty in future
models.
Reactive Transport Modeling for the Proposed Dewey Burdock Uranium In-Situ Recovery Mine,
Edgemont, South Dakota, USA (Johnson 2011) was presented at the IMWA 11* International
Mine Water Association Congress, "Mine Water - Managing the Challenges." This paper
Draft Technical Report 90 Revised Draft - November 26, 2012
-------
provides an initial reactive-transport simulation, which supports a conceptual understanding of
uranium roll-front formation, current ground water conditions, mining geochemistry, restoration
geochemistry, and long-term ground water quality at a uranium ISR site. As the author states,
this is the starting point for additional refinements to improve the model.
The reactive-transport modeling discussed in the paper used PHAST (Parkhurst et al. 2010), a
relatively simple ground water flow code coupled with PHREEQC (Parkhurst and Appelo 1999),
to calculate geochemical conditions at each time step. For this work, generic values were
assigned to the ground water flow velocities and mass balances for the solid phase. Site-specific
data are expected to be added in the future. Even so, some simulations were conducted to
simulate pre-mining conditions. In one simulation, ground water with no dissolved oxygen was
transported through the model domain. The result was a solid phase uraninite roll-front
associated with pyrite on the solid-phase reduced side; uranium and dissolved oxygen were not
found in solution.
Uranium in-situ mining was simulated using a leach solution fortified with oxygen and carbon
dioxide. The resulting oxidation made uranium soluble, and the carbon dioxide created a
complexing agent. A five-spot well pattern was simulated with a center pumping well and four
surrounding injection wells. In the ore zone, the result was elevated concentrations of uranium in
the ground water where the ore zone was being mined.
During the restoration phase, the existing wellfield was used to flush out the mining solutions
from the ground water. This process was simulated as water with low dissolved constituent
concentrations, but with 50 parts per billion (ppb) residual uranium. In one simulation, oxygen
was left in the restoration fluids, and in another simulation, oxygen was kept at zero. At this
stage, reductant addition to help precipitate uranium could be simulated, but that simulation was
not completed for this paper. As noted, the model will be refined as site-specific data are
incorporated.
Ongoing Research
To assist decisionmaking about ISR design, operations, restoration, closure and monitoring will come
from the application of reactive transport models. Many such models are already in use to support the
ISR mining and new developments in this field (e.g., PHREEQC, Geochemist's Workbench®),
although there are a few sophisticated codes that allow the reactive transport processes to be coupled
to the ground water flow regime, such as PHAST or PHT3D. To further investigate data needs, data
collection, model applicability and potential modeling approaches, the EPA has entered into a
corporative agreement with the U.S. Geological Survey under a Regional Applied Research Effort
(RARE).
4.7.4 Demonstrating Long-term Stability of Restored ISR Wellfields - Long-Term
Monitoring and Geochemical Modeling
After wellfield restoration efforts have stopped, the restored wellfield is typically monitored for
periods of six months to several years, with the intent of demonstrating that the ground water
chemistry in the wellfield has reached a "steady state" at compositions as close as possible to the
pre-mining background (baseline) levels. While this post-restoration monitoring will provide
Draft Technical Report 91 Revised Draft - November 26, 2012
-------
some confidence that the pre-mining geochemical conditions in the wellfield have returned, the
situation may not persist over the long term. The initial injection of chemicals to oxidize and
mobilize the uranium ore will most likely also oxidize the chemical reducing agents (electron
acceptors) originally present in the aquifer that acted to sequester the uranium in the ore body.
These materials are thought to consist of iron sulfide minerals and organic material. If the
amounts of these reducing agents have been decreased sufficiently by the mining process, the
continual movement of oxidizing waters into the ore zone from up gradient may be too strong for
the remnants of the reducing agents to maintain chemically reducing conditions. In that case, the
uranium chemically reduced by the restoration process will be re-oxidized and migrate down
gradient out of the wellfield over time. In such a situation, the restoration efforts are simply
palliative, since they may not restore the underlying mechanism(s) responsible for sequestering
the uranium. The long-term prospect is that radionuclides and perhaps toxic metals will move out
of the depleted ore zone and into surrounding aquifer areas, because the underlying mechanism
for sequestering them has been weakened or totally removed.
To detect this possibility, two approaches are possible. In one approach, the post-restoration
monitoring could be continued for periods that may be tens of years in duration to detect any
deterioration of the chemically reducing conditions in the wellfield, at which point, if detected,
remediation efforts could be developed. Such a long monitoring process is necessary, because
the kinetics of a re-oxidation process in the field can be slow. Section 7.7.2 examines statistical
measures for determining stability in a sequence of measurements considering the influence of
natural variability variations and rates of change in the measured parameter. Table 7-20 and
Table 7-21 show how many samples (assuming quarterly sampling) are needed to attain high
levels of confidence (95% and 99%) that a trend can be detected for various assumed levels of
natural variability and parameter change rates. For relatively low natural variability levels and
rates of change, the number of samples, and consequently the post-restoration monitoring period,
significantly exceed typical post-restoration sampling periods used in practice. A 30-year post-
restoration monitoring period proposed in the rulemaking establishes a monitoring period
consistent with RCRA regulations.
Another alternative is to use geochemical modeling to demonstrate that the restored wellfield and
the down gradient geochemical conditions in the exempted aquifer area are sufficient to maintain
chemically reducing conditions over the long-term. The modeling used to address the post-
restoration possibilities involves relatively simple aqueous speciation modeling, as well as more
complex process modeling and perhaps coupled flow and contaminant transport models. As
described above, these types of models are readily available and typically used to design the
wellfield geometry and optimize production, as well as help design restoration efforts. Their use
in assessing post-restoration behavior at an ISR site is discussed further below. The intent of the
modeling from a regulatory perspective is to answer two questions:
• Will the ground water chemistry in the restored wellfield keep uranium and other
mobilized toxic metals in place?
• Does the down gradient portion of the exempted aquifer have the reducing capacity to
remove uranium from ground waters that migrate into the area from the up gradient
wellfield area?
Draft Technical Report 92 Revised Draft - November 26, 2012
-------
As a first step in geochemical modeling, it is important to assess the chemical reducing capacity
of the ore zone and down gradient portion of the exempted aquifer before the mining is initiated
and after the restoration efforts have ceased. This requires information on the nature and amount
of chemical reducing agents present in the ore zone before and after the ISR process, as well as
down gradient of the ore zone in the remainder of the exempted aquifer area. Concentrations of
pyrite (and other iron sulfide minerals) and organic material, as well as other electron acceptor
species in the ore zone, should be measured before ISR operations begin and after restoration
efforts end. Aqueous indicators of chemically reducing conditions (such as various redox
sensitive couples, e.g., sulfide/sulfate, ferric/ferrous iron, redox state measurements, other
oxidizing agents such as dissolved oxygen, and nitrate levels) should also be measured in the
ground waters. Information on solid-phase electron acceptors can be gathered from analyses of
cores taken within the exempted aquifer before operations begin and after restoration efforts
stop. Section three of this document contains tabulations of aqueous species that should be
monitored for geochemical modeling applications.
With the information on concentrations of chemically reducing agents in the ground water and
solids within the wellfield and in the down gradient portion of the exempted aquifer, mass
balance calculations and process modeling exercises can be done to determine if the reducing
capacity of the restored wellfield and down gradient area is sufficient to keep the uranium
remaining after the mining in a reduced state, minimizing the potential for migration. While such
modeling evaluates the potential for maintaining chemically reducing conditions, kinetic data
would allow more sophisticated modeling of reaction progress within the exempted aquifer,
including coupling of reaction path and ground water flow modeling. This modeling would
examine the reaction rates and evolution of changes in aqueous chemistry as ground waters
bearing oxygen and uranium (VI) enter the wellfield area from the up gradient direction, interact
with the "restored" wellfield chemistry and move down gradient into the un-mined portion of the
exempted aquifer and eventually into the non-exempt portion of the aquifer. Kinetic data to
support such modeling is limited (largely to laboratory data with limited field data), but the
research described above should significantly increase the field database and understanding of
these processes.
With sufficiently robust geochemical modeling demonstrating that the system can maintain
chemically reducing conditions over the long-term and contain any uranium migrating out of the
restored wellfield area within the down gradient portion of the exempted aquifer, the appropriate
regulatory authorities may allow a shorter than 30-year post-restoration modeling period. This
modeling demonstration is also important for regulatory decisions about the need for alternative
concentration levels and the protection of ground water resources in the down gradient area
outside the boundary of the exempted aquifer.
Draft Technical Report 93 Revised Draft - November 26, 2012
-------
5.0 ACTIVE/EXISTING ISR FACILITIES: MONITORING ISSUES
The standards in 40 CFR 192.32 refer to several sections of RCRA 40 CFR Part 264, Subpart F,
which describe EPA's regulatory approach for releases to ground water from waste management
units that store, treat, and dispose of hazardous waste. Although §264.97 is not specifically cited
in §192.32, it provides some useful guidance regarding general requirements that could be
considered for establishing a suitable ground water baseline:
(g) In detection monitoring or where appropriate in compliance monitoring, data
on each hazardous constituent specified in the permit will be collected from
background wells and wells at the compliance point(s). The number and kinds of
samples collected to establish background shall be appropriate for the form of
statistical test employed, following generally accepted statistical principles. The
sample size shall be as large as necessary to ensure with reasonable confidence
that a contaminant release to ground water from a facility will be detected. The
owner or operator will determine an appropriate sampling procedure and
interval for each hazardous constituent listed in the facility permit which shall be
specified in the unit permit upon approval by the Regional Administrator. This
sampling procedure shall be:
(1) A sequence of at least four samples, taken at an interval that assures, to
the greatest extent technically feasible, that an independent sample is
obtained, by reference to the uppermost aquifer's effective porosity, hydraulic
conductivity, and hydraulic gradient, and the fate and transport
characteristics of the potential contaminants, or
(2) An alternate sampling procedure proposed by the owner or operator and
approved by the Regional Administrator.
In practice, the procedures for establishing the ground water baseline are site
specific and are included in the facility license issued by the NRC or Agreement
State.
5.1 Ground Water Baseline: Case Studies
The requirements for baseline monitoring vary from state to state. (See Section 3.5 for additional
details.) In Texas, 26 chemical constituents are measured before mining to establish a baseline,
as shown in Table 5-1. This is example data from Production Authorization Area (PAA) No. 1 at
the Zamzow ISR facility. Baseline values represent the highest average concentration from either
the production or mine area, and are commonly selected as initial restoration goals (Hall 2009).
In Table 5-1, the mine area is defined by a line through a ring of monitor wells in the production
zone and the production area is defined by a line generally through the outer perimeter of
injection and recovery wells.10 TCEQ regulations require a minimum of five baseline wells or
TCEQ Chapter 331, Subchapter A, §331.2.
Draft Technical Report
94
Revised Draft - November 26, 2012
-------
one well per four acres, whichever is greater (TCEQ Title 30, Rule §331.104). Sampling
frequency and number of samples are not specified by the TCEQ regulations.
Uranium Energy Corporation filed a production area authorization (PAA-1) application for the
Goliad Uranium Project with the TCEQ on August 27, 2008, and this application was amended
on March 27, 2009, to include data from additional baseline wells (Sass 2011). The amended
application included 18 baseline wells from the 36-acre production area (i.e., 0.5 wells/acre). The
average uranium content for these 18 wells was used to establish a baseline uranium value of
0.115 mg/L. The initial sampling of the 18 wells was conducted over a period of about
11 months (4 wells about July 2007, 6 wells in April 2008, and 8 wells in July 2008). The
amended PAA-1 application did not include data from subsequent sampling of all 18 wells in
July 2009 and again in November 2009. The averaged uranium assay results for these subsequent
periods were 0.029 mg/L and 0.005 mg/L, suggesting that the initial samples were not indicative
of geochemical equilibrium. Had the full time series of samples been included in the PAA-1
application, a more rigorous baseline standard would have been set against which to measure
restoration.
Table 5-1. Baseline Water Quality Data for Zamzow PAA-1
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
Parameter
Cadmium
Magnesium
Sodium
Potassium
Carbonate
Bicarbonate
Sulfate
Chloride
Fluoride
Nitrate - N
Silica
pH
TDS
Conductivity
Alkalinity
Arsenic
Cadmium
Iron
Lead
Manganese
Mercury
Selenium
Ammonia
Uranium
Molybdenum
Radium-226
Unit
mg/1
mg/1
mg/1
mg/1
mg/1
mg/1
mg/1
mg/1
mg/1
mg/1
mg/1
Std. units
mg/1
umhos
Std. units
mg/1
mg/1
mg/1
mg/1
mg/1
mg/1
mg/1
mg/1
mg/1
mg/1
pCi/1
Production Zone
Mine Area**
Low
122
15
239
19
0
128
454
350
0.16
0.01
31
6.6
1,697
2,720
105
0.001
0.0001
0.01
0.001
0.009
0.0001
0.001
0.01
0.001
0.001
1.5
Average
317
38.4
387
30.3
0
297
793
503
0.54
0.16
51.6
7.0
2,289
3,204
275
0.009
0.001
0.915
0.001
0.224
0.0004
0.01
0.374
0.171
0.03
155
High
552
84.2
750
49
0
400
1,520
936
1.19
0.9
85
7.66
3,220
4,300
400
0.03
0.007
8.0
0.006
0.82
0.0018
0.01
1.4
1.7
0.95
959
Production Area
Low
195
3.0
235
18.9
0
157
441
394
0.01
0.01
11
6.68
1,810
2,680
206
0.001
0.0004
0.03
0.001
0.01
0.0001
0.001
0.01
0.001
0.001
6.5
Average
269
21.1
383
26.7
0
269
601
538
0.36
0.14
43.9
7.0
2,037
3,049
238
0.006
0.001
0.075
0.004
0.118
0.0006
0.004
0.298
0.039
0.226
152
High
390
40
466
90
0
346
940
662
0.50
0.49
74
7.45
2,360
3,430
204
0.044
0.0013
0.26
0.02
0.19
0.001
0.01
0.78
0.432
2.1
744
** Monitor wells
Source: Hall 2009
Draft Technical Report
95
Revised Draft - November 26, 2012
-------
In its license application for the Moore Ranch Uranium Project in Campbell County, Wyoming,
Energy Metals Corporation (2007) proposed to establish the wellfield baseline by sampling
production zone wells four times, with a minimum of 2 weeks between samplings (NRC 2010,
Section 6.3.1.1). Energy Metals also proposed to sample one well for each 3 acres of mine unit.
Data for each sampled parameter would be averaged and used to calculate restoration goals. The
average and range of baseline values in the production zone would then be used to assess the
effectiveness of subsequent ground water restoration.
In another example, at Mine Unit 4 of the Christensen Ranch Project in Wyoming, the wellfield
covered about 12 acres and, consequently, 12 injection or production wells were used to establish
baseline ground water conditions within the ore zone, which in turn set the restoration goals
(COGEMA 1994). The number of baseline wells was based on NRC guidance in NUREG-1569
(NRC 2003, p. 5-39) of one well per acre.
Commercial-scale uranium ISR facilities usually have more than one wellfield. For example, the
Crow Butte facility in Dawes County, Nebraska, has constructed 10 wellfields since 1991 (Crow
Butte 2007). The locations and boundaries for each wellfield are adjusted as more detailed data
on the subsurface stratigraphy and distribution of uranium mineralization are collected during
wellfield construction.
5.2 Wellfield Restoration
Wellfield restoration is defined as those actions taken to ensure that the uranium extraction
process will not adversely affect the quality of the ground water adjacent to the ISR wellfields
(NRC 2001). This requires returning the wellfield water quality parameters to meet the
restoration goals included in the facility license issued by NRC or the Agreement State. Based on
pre-mining monitoring, the operator establishes baseline values for the ground water quality. The
regulator then uses these baseline values to set restoration goals in the wellfield license.
The portion of the aquifer undergoing uranium extraction is exempt from EPA regulatory
protection under the SDWA (specifically the UIC Program at 40 CFR Part 144). However,
ground water adjacent to the exempted portion of the aquifer must still be protected, and ground
water protection provisions for this water are in effect. Similar to the NRC Agreement State
provisions,11 an EPA Primacy State may impose more stringent requirements for ground water
restoration than the federal program (NRC 2003). Ground water restoration requirements may
vary from state to state. Of particular importance is underground injection and point source
discharge into surface waters. Currently, UIC programs are administered (as authorized by EPA)
in Wyoming, Nebraska, and New Mexico. South Dakota administers the program jointly with
EPA.
It should be noted that UMTRCA gives NRC the authority to require licensees to restore
ground water in the mined aquifer, and EPA to set the standards for that restoration. UMTRCA
11 Texas, Colorado, and Utah operate as Agreement States under NRC regulations in establishing state-
specific ISL regulations, while facilities in Wyoming, New Mexico, and South Dakota are directly regulated by
NRC. Nebraska is also an Agreement State, but since it does not have specific ISL regulations, its facilities are
regulated by NRC.
Draft Technical Report 96 Revised Draft - November 26, 2012
-------
operates separately from SDWA UIC rules, and requires restoration of the ground water in the
mined aquifer, even though the water may not be considered potable under SDWA. EPA's UIC
regulations provide protections for non-exempt portions of the aquifer, but generally do not
require restoration of the exempt portion unless it may affect drinking water outside its
boundaries. State rules under UIC delegation are generally weaker than EPA's and NRC's
UMTRCA requirements.
5.3 Wellfield Restoration: Case Study
Table 5-2 summarizes restoration results from 22 PAAs in Texas (Hall 2009). It is apparent that,
for all of the PAAs, post-restoration analyses exceeded the baseline for some of the parameters
tested. Similar information on restoration of sites in other states was extracted from NRC 2009
and is included as Attachment C. Table 5-2 also shows that all of the post-restoration parameters
exceeded the baseline in some wellfields. This illustrates the difficulty of restoring wellfields to
baseline conditions. However, for most of the species with quantitative limits set by MCLs or
secondary drinking water standards, the quantitative limits were below the baseline. Exceptions
were fluoride, nitrate, and sulfate where the baseline limits were below recommended standards
for all 22 PAAs.
Draft Technical Report 97 Revised Draft - November 26, 2012
-------
Table 5-2. Ground Water Chemistry of Texas In-Situ Uranium Production
Authorization Areas
Analyte
EPA and TCEQ
Drinking Water
Standards
(mg/1)
Baseline Range
Post-restoration
Range
PAAs with
Baseline Above
MCL or
Recommended
Standards
PAAs with Post-
restoration
Water Above
MCL or
Recommended
Standards
PAAs Where
Post-
restoration
Analyses
Exceed
Baseline
PAAs Where
Post-
restoration
Analyses are
Below
Baseline
EPA and TCEQ Primary Maximum Contaminant Levels (MCLs):
Arsenic
Cadmium
Fluoride
Lead
Mercury
Nitrate
Selenium
Radium (226 and
228 Ra: pCi/1)
Uranium
0.01
0.005
4
0.02
0.002
10
0.05
5 pCi/1
0.03
.004-0.23
0.0001-0.0126
0.21-1.8
0.003-1.97
0.0001-0.445
0.031-10.0
0.001-0.049
9.36-429.8
0.025-2.0
.002-.323
0.0001-0.01
0.29-1.6
0.001-0.05
0.0001-0.01
0.001-2.8
0.001-0.102
5.2-149
0.013-3.02
77%
45%
0%
81%
9%
0%
18%
100%
95%
55%
23%
0%
18%
0%
0%
4%
100%
86%
18%
27%
31%
9%
22%
4%
54%
4%
68%
82%
73%
69%
91%
64%
96%
45%
96%
32%
TCEQ Secondary Recommended Standards:
Sulfate
Chloride
Total Dissolved
Solids
Iron
Manganese
300
300
1000
0.3
0.05
15.8-250
196.9-3505
785.7-6349
0.04-5.49
0.01-0.41
78-3881
138-3326
706.3-6155
0.01-2.7
0.01-0.84
0%
86%
81%
54%
77%
18%
86%
77%
9%
50%
86%
22%
31%
4%
40%
14%
78%
55%
96%
60%
No Established MCL or Secondary Standards
Calcium
Magnesium
Sodium
Potassium
Carbonate
Bicarbonate
Silica
Conductivity
(umhos/cm)
Alkalinity (as
CaCO3)
Molybdenum
Ammonia-N
-
-
-
-
-
-
-
-
-
-
-
4.13-241
0.477-125
200-2356
6.38-101
0.1-17.9
160-500
16.3-76
1310-11160
134-349
0.01-0.2
0.01-7.49
14.7-191
2.27-53
169-2247
6.1-70
0-14.6
160-500
13.4-77.6
1429-3697
145-408
0.0001-3.38
0.04-120
77%
72%
31%
14%
50%
66%
19%
76%
81%
42%
76%
23%
28%
65%
86%
30%
25%
81%
24%
10%
54%
24%
Baseline and post-restoration data were available for all 22 PAAs with the exception of Ra, Mo, K, Si, Bicarbonate, Ammonia (21), Conductivity
(14), Alkalinity (!!),& Carbonate (10)
Source: Hall 2009
Draft Technical Report
98
Revised Draft - November 26, 2012
-------
6.0 ISSUES ASSOCIATED WITH ESTABLISHMENT OF POST-RESTORATION
STEADY STATE
During restoration, the operator monitors progress by periodic sampling of the ground water
constituents until steady-state conditions are attained. Establishment of steady state requires that
the ground water potentiometric surface be restored, to the extent practicable, to its preleaching
status, so that the flow regime is similar to that existing before mining. In addition, constituents
in the ground water must be in compliance with restoration goals and remain at those levels for a
sufficient period to demonstrate that the results are not trending upwards to higher concentration
levels. EPA describes a "steady state" as having the following relevant components (EPA 1992,
Chapter 7):
La After treatment, the water levels and water flow, and the
corresponding variability associated with these parameters (e.g., seasonal
patterns), should be essentially the same as for those from comparable
periods of time prior to the remediation effort.
or
1. b. In cases where the treatment technology has resulted in permanent
changes in the ground water system, such as the placement of slurry wells, the
hydrologic conditions may not return to their previous state. Nevertheless,
they should achieve a state of stability which is likely to reflect future
conditions expected at the site. For this steady state, the residual effects of the
treatment will be small compared to seasonal changes.
2. The pollutant levels should have statistical characteristics (e.g., a mean and
standard deviation), which will be similar to those of future periods.
The first of these components addresses the general behavior and characteristics of the ground
water at the site. The second is more judgmental and requires projection of future contamination,
based on available current information. These projections cannot be made with certainty;
however, various criteria can be used in determining whether a steady state has been reached.
Section 7.8 of this report discusses statistical tests for measuring attainment of steady state.
When the regulator is satisfied that steady state has been achieved, the operator is authorized to
undertake long-term post-restoration stability monitoring.
6.1 Post-restoration Stability Monitoring
Once the operator concludes that restoration has been completed and has obtained concurrence
from the regulator(s) that a steady state has been established, post-restoration stability
monitoring begins. The purpose of the stability monitoring is to demonstrate that the aquifer
conditions established at the end of restoration are sustainable over time. Currently, the duration
of stability monitoring is a site-specific period of time established in the license(s). In the past,
the license-established restoration period typically has been about 6 months (see case histories in
Attachment B). More recently, the trend has been to increase the monitoring period established
in the license. In practice, the actual period of stabilization may be several years, based on
Draft Technical Report 99 Revised Draft - November 26, 2012
-------
iterative analyses of additional samples requested by the regulators see Table 6-1). If the
sandstone in the aquifer is heterogeneous, extended restoration times may be required to ensure
that ground water in slow pathways is addressed.
Table 6-1. Post-restoration and Stability Monitoring Periods
Facility
Name
Moore Ranch
Wellfield 1
Moore Ranch
Wellfield 2
Nichols
Ranch
Lost Creek
Ruth Test Site
State
Campbell County,
Wyoming
Campbell County,
Wyoming
Campbell and Johnson
Counties, Wyoming
Sweetwater, Wyoming
Johnson, Wyoming
Projected or
Estimated
Restoration
Period
3. 5 years
5.25 years
1 to 5 years
2 years
12 months
Projected or Estimated
Post-restoration
Monitoring Period
1 year (quarterly)
1 year (quarterly)
1 year (quarterly)
6 months (monthly)
12 months
Comment/Reference
NUREG-1910
Supplement 1
NUREG-1910
Supplement 1
NUREG-1910
Supplement 2
NUREG-1910
Supplement 3
Schmidt 1989
6.2 Factors That Affect Time Frames for Post-mining Monitoring
This section summarizes the factors that must be understood to determine when the impacted
aquifer has reached a steady-state condition.
6.2.1 Fate and Transport Processes
The monitored time frame is dependent on mass-balance estimates of how much extraction fluid
remains in the aquifer. Monitoring during operations needs to determine a mass balance of the
total volume of lixiviant injected into the system and the volume withdrawn. The lixiviant used
to extract the uranium can mask baseline constituents and affect reaction kinetics. Knowing how
much lixiviant remains in the aquifer will aid in understanding whether some reactants are still in
the system, have migrated outside the monitored area, or have been temporarily sequestered in
low-permeability zones, or are undergoing incomplete or slow reactions that may later release
constituents.
6.2.1.1 Specicttion
The environmental chemistry of uranium is largely dictated by its oxidation state
(e.g., Fanghanel and Neck 2002). Under ambient oxidizing conditions, the predominant uranium
oxidation state is U(VI). Where oxygen is limited, U(IV) may dominate. The metallic form,
U(0), does not occur naturally, and is readily oxidized to U(IV) and eventually U(VI), upon
exposure to oxidizing conditions. The mechanisms for the oxidation of U(0) and U(IV) to U(VI)
are well established (e.g., NRC 2007). It is rare to find other oxidation states of uranium [e.g.,
U(V) and U(III)] under natural conditions, due to their instability. However, stable U(V) has
been found on mica surfaces (Ilton et al. 2004, Ilton et al. 2005, Ilton et al. 2008).
Draft Technical Report
100
Revised Draft - November 26, 2012
-------
In general, the solubility and therefore the mobility of uranium are greatest when it is in the
U(VI) state. Complexation of U(VI) by inorganic anions, such as carbonate, fluoride, and
phosphate, may enhance the solubility and mobility of this species. When reducing conditions
are present, U(IV) is generally immobile and found either as an insoluble oxide (uraninite) or a
silicate (coffinite). Under oxidizing conditions and near neutral pHs, U(VI) species dominate
aqueous uranium concentrations. These highly soluble species are generally either hydroxy or
carbonate complexes of the uranyl (UC>22+) cation, although elevated concentrations of potential
inorganic ligands near the ISR target zone may exert greater influence on U(VI) speciation (e.g.,
phosphate).
A detailed discussion provided by Demuth and Schramke (2006) describes the geochemical
processes controlling the fate and transport of uranium, radium, selenium, arsenic, molybdenum,
sulfate, iron and manganese. The authors summarize their findings by reiterating the importance
of redox conditions on influencing the mobility of uranium, selenium, arsenic, molybdenum, and
sulfur. Although these constituents are likely to be more mobile under relatively oxidizing
conditions, adsorption and desorption reactions between uranium, selenium, arsenic, and
molybdenum with iron oxide surfaces are particularly important controlling reactions, because
iron oxides are widespread in the hydrogeologic environment as coatings on other solids. The
precipitation of sulfate phases (e.g., barite, celestite, gypsum) is likely to attenuate sulfate and
radium-226 by solid solution. Radium-226 is also strongly attenuated by adsorption onto clay
minerals. Under oxidizing conditions, iron and manganese tend to form relatively immobile iron
oxyhydroxides and manganese oxides. These oxides may provide adsorption sites for many trace
metals, such as uranium, arsenic and molybdenum. Under reducing conditions, particularly as
sulfate is consumed and the sulfur is converted to sulfide, concentrations of dissolved metals
such as molybdenum decrease as solid-phase metal sulfides are formed.
Calcium (or other alkaline earth metals, such as magnesium) and inorganic carbon in ground
water tend to dominate the aqueous speciation of U(VI) under near neutral pH conditions. The
presence of these species is common in many natural ground water systems (Hem 1985), and as
noted below, these speciation characteristics also influence the degree to which U(VI) will
adsorb onto aquifer solids. Under reducing conditions, U(IV) species, primarily the uranyl cation
and its complexes, predominate, but because of the very low solubility of U(IV) minerals, reach
maximum concentrations on the order of 10 nanomolar (2.4 micrograms U/L). For all practical
purposes, therefore, only U(VI) aqueous species are at sufficient concentrations to be of
environmental concern.
Chemical reaction kinetic equations or equilibrium thermodynamic equations can be used to
describe chemical interactions among dissolved chemical species, the dissolution of immobile
solid phases, or the formation and precipitation of new, immobile solid phases. Geochemical
modeling is often performed at ISR facilities to gain a better understanding of
thermodynamically controlled processes that include mineral dissolution/precipitation,
oxidation/reduction, and adsorption/desorption.
Most of the available computer codes assume thermodynamic equilibrium and do not have a
method of calculating reaction rates (i.e., kinetics). If a mineral forms or dissolves slowly in a
system, the model developed from these codes will not account for these kinetic effects. This is
Draft Technical Report 101 Revised Draft - November 26, 2012
-------
not a major limitation for most aquifer systems, where residence times are measured in years;
however, kinetic effects can become more important in modeling reactions anticipated to occur
during applied remediation methods, such as the injection of reactants into an aquifer. Ground
water flow rates in aquifers undergoing ISR are typically slow, on the order of feet to tens of feet
per year, suggesting that the development of geochemical models for these environments may
not be severly limited due to a lack of site-specific knowledge of kinetic effects for chemical
reactions that occur during and after restoration efforts.
6.2.1.2 Speciation: Case Study
The experience with iron at the Crow Butte ISR facility is illustrative of speciation problems.
Crow Butte Resources (CBR) experienced difficulty in reaching desired iron levels during
wellfield restoration. During the initial stabilization monitoring period in 1999, the iron
concentration averaged 0.089 mg/L. Subsequent testing in the summer of 2002 showed an
average iron content of 0.278 mg/L. The operator attributed this to speciation initiated by the
original injection of lixiviant, with subsequent transitory solubility increases resulting from the
selected restoration method. As CBR stated (Crow Butte 2002):
CBR believes that the elevated iron concentrations are due to the restoration
process and will ultimately decrease to concentrations well below the restoration
standard. During the in situ mining process, when the groundwater is oxygenated
and the Eh is positive, the iron contained in pyrites is oxidized to ferric iron and
forms ferric oxyhydr oxides. The ferric oxyhydroxides are extremely insoluble,
which explains the very low concentrations of iron in solution during mining,
indicated by the end of mining values which, with the exception of one restoration
well (PR-19), were below the detection limit of 0.05 mg/L. During the active
restoration process, however, sodium sulfide is used as a reductant to decrease
the Eh of the groundwater. As the Eh drops, the stable solid iron phase is reduced
from ferric iron to ferrous iron, which is more soluble. During the transition from
ferric to ferrous iron, the iron concentration in the groundwater increases
significantly. This increase in the iron concentration is transitory and, as the Eh
continues to decrease, iron sulfide minerals will be the dominant iron phase.
Because of the relative insolubility of these iron sulfide minerals, this will cause a
significant decrease in the iron concentration in solution. Based on these
mechanisms, CBR expects that the elevated concentrations of iron at the current
time will ultimately decrease.
Without greater insight into the mineralogy and flow and transport processes within the aquifer,
however, there is no reliable means to test CBR's hypothesis. Active research is being conducted
to better understand the processes controlling aquifer restoration. For example, Cameco
Resources is currently seeking permission from the Wyoming Department of Environmental
Quality to conduct a series of aquifer restoration experiments at Smith Ranch mine, which
include: (1) tracer tests to determine the hydrologic pathways between injector and recovery
wells; (2) bio-stimulation tests to determine the viability of using naturally occurring bacteria to
re-precipitate uranium and other redox-sensitive species; and (3) natural attenuation tests to
Draft Technical Report 102 Revised Draft - November 26, 2012
-------
determine the ability of the aquifer down gradient of the mining zone to immobilize
contaminants (Cameco 2012).
6.2.1.3 Solubility
Under most natural conditions, the thermodynamically stable uranium solid phases will be either
U(VI) or U(IV) compounds. The most stable U(VI) compounds are the phosphates and
vanadates, but their formation is often limited by the relatively low concentrations of these two
anions, and thus more soluble U(VI) oxides, such as schoepite, are often seen if any U(VI) solid
phases are present. A significant fraction of the solid-phase U(VI) will be adsorbed to iron
(hydr)oxide surfaces, to the edges of clay minerals, and to organic matter, rather than
precipitated as discrete uranium phases. Maximum solubility of uranium is seen in oxidizing,
phosphate-free, carbonate-rich solutions, and consequently, carbonates (or bicarbonates) and
oxygen or hydrogen peroxide are the principal reagents used for ISR mining.
Under reducing conditions, the stable U(IV) solid phases are uraninite and, if high amounts of
dissolved silica are present, coffinite. Organic complexes of U(IV) associated with humic
material may also retain U(IV) in the solid phase. The solubility of the U(IV) phases is extremely
low, and thus the presence of reducing conditions effectively halts or slows the movement of
uranium in soils and sediments, provided that colloidal-sized uranium-bearing particles are not
formed and transported. The most common uranium ore-forming process involves reductive
precipitation of U(IV) phases as a result of microbiological activity to form a roll-front deposit
(Langmuir 1997). The stability fields for U(VI) and U(IV) as a function of pH and Eh for various
water compositions suggest that a wide variety of uranium-bearing precipitates are possible,
especially in complex ground water systems that invariably contain silica, carbonate/bicarbonate,
calcium/magnesium, sodium, and sometimes phosphate. Furthermore, it may be difficult to
predict associations of uranium in the solid phase based on analysis of aqueous chemical data
and solubility predictions from thermodynamic chemical data. In the absence of confirmatory
solid-phase characterization data, equilibrium model projections indicate only the possible
formation of specific uranium-bearing precipitates.
6.2.2 Natural Attenuation Processes
Natural attenuation processes include a variety of physical, chemical, and biological processes
that can act to reduce the mass, mobility, volume, or concentration of contaminants in ground
water. Attenuation processes important at ISR sites include pH buffering and acid neutralization,
adsorption at the mineral-water interface, mineral precipitation, and dilution/dispersion.
6.2.2.1 Adsorption
Adsorption processes are typically categorized by the relative "strength" of the interaction
between the adsorbate (species in solution) and the surface or adsorbent. If water molecules are
positioned between the cation or anion and the surface, the adsorption complex is referred to as
outer sphere and is considered to be weak. Conversely, if upon adsorption, the adsorbate loses
waters of hydration such that no water molecules are positioned between the cation or anion and
the surface, the adsorption complex is referred to as inner sphere and is considered to be strong.
Draft Technical Report 103 Revised Draft - November 26, 2012
-------
Adsorption of uranium typically involves inner-sphere complexation of uranyl (i.e., UC>22+)
species by oxygen ligands at the surfaces of iron oxyhydroxides, phosphates, and layered
silicates. Uranyl species exhibit a high affinity for iron oxyhydroxide surfaces and for both basal
and edge sites on layered aluminosilicates, such as the clays smectite and vermiculite.
Adsorption of U(VI) to the aluminosilicate mineral, muscovite, has been observed in aquifer
sediments at the Hanford Site in Richland, Washington (McKinley et al. 2007). Complexation of
U(VI) by organic ligands in solid humic materials (primarily carboxylic-acid and phenolic
groups) may also serve to remove uranium in shallow ground water systems (Sowder et al.
2003).
A compilation of published Kd values for U(VI) sorption onto soils/sediments is documented in
EPA 1999a. However, the authors of that compilation recognized that there are major limitations
to the application of published KdS for site-specific applications where either the ground water
chemistry or the aquifer matrix differs significantly from the conditions under which a Kd was
determined (Ochs et al. 2006). One interesting study examined uranium sorption in a sandstone
aquifer under site-specific conditions by back-calculating a Kd for the observed uranium
distribution after developing a validated ground water flow model for the aquifer using flow rates
and carbon isotope data for the ground water system (Pearson et al. 1983). Areas of chemically
oxidizing and reducing conditions in the aquifer were observed, indicating that uranium
reduction processes were active similar to those involved in the formation of roll-front deposits.
A Kd of 6 was proposed for uranium sorpion on the ssandstones based on the field data and
validated hydrologic model. Most of the information on uranium sorption reported in the
literature is derived from laboratory measurements, in contrast to the the study referenced above.
Davis et al. (2004) document a Surface Complexation Model (SCM) alternative to the constant K
approach. As described by Davis et al. (2004), there are two major methods for applying the
SCM concept; the Component Activity (CA) and the Generalized Composite (GC) approaches.
In the CA approach, it is assumed that a mineral assemblage is composed of a mixture of one or
more reference phases, whose surface chemical reactions are known from independent studies of
each phase. Next, based on a measurement of the relative amounts or surface areas of each
mineral present in the soil or sediment, adsorption by the mixture of phases can be predicted by
an equilibrium calculation, without any fitting of experimental data for the mixture. In the GC
approach, the surface of the mineral assemblage is considered too complex to be quantified in
terms of the contributions of individual phases to adsorption. In the GC approach, it is assumed
that adsorption can be described by mass laws written with "generic" surface functional groups,
with the stoichiometry and formation constants for each mass law determined by fitting
experimental data for the mineral assemblage as a whole.
This SCM approach incorporates the important influence of uranium solution speciation, while
avoiding the need to model the influence of individual mineral components (and their respective
surface charging behavior). While this approach still requires site-specific data, it provides a
means for projecting the influence of changes in ground water chemistry on uranium sorption.
The chemistry of ground water may be influenced by reaction with aquifer solids and/or external
recharge/infiltration from atmospheric precipitation or surface water. As previously noted,
alkalinity influences the aqueous speciation of U(VI), and it also influences the degree of
sorption of U(VI) onto iron oxyhydroxides and aquifer solids in which these minerals control
Draft Technical Report 104 Revised Draft - November 26, 2012
-------
uranium partitioning (e.g., Um et al. 2007). It has been demonstrated that changes in ground
water chemistry influence the transport of U(VI) through an aquifer (Yabusaki et al. 2008).
Alternatively, transition from oxidizing to reducing conditions along the transport pathway may
be accompanied by a shift from adsorption of U(VI) species to precipitation of U(IV)-bearing
solids (Davis et al. 2006). Reactive-transport models used to project subsurface uranium mobility
directly incorporate the influence of major ion chemistry and redox conditions on the chemical
speciation of uranium.
Field evidence shows that adsorption of uranium to mineral surfaces within an aquifer may be an
intermediate step to the formation of uranium-bearing precipitates. Murakami et al. (2005) have
observed the association of nanoparticulate U(VI)-phosphate precipitates with iron
oxyhydroxides in the weathering zone down gradient from a uranium ore deposit. The U(VI)
mineral was identified as metatorbernite, which was present in ground water that was
undersaturated with respect to precipitation of this mineral. Characterization of the textural
associations between the nanocrystalline metatorbernite and iron oxyhydroxides present as
fissure fillings, clay coatings, and nodules, along with compositional relationships between
copper, phosphorus, and uranium (Sato et al. 1997), indicated that the formation of uranium
precipitates was a secondary step following initial adsorption of these constituents onto iron
oxyhydroxide mineral surfaces (Murakami et al. 2005). As summarized by Payne and Airey
(2006), the observations in this subsurface system provide a point of reference for designing site
characterization strategies and developing both conceptual and analytical models for interpreting
and projecting uranium mobility in ground water.
O'Loughlin et al. (2003) believe that mixed ferrous/ferric hydroxides (i.e., green rust) play a
central role in the biogeochemistry of iron. The authors conclude that their experimental results
clearly indicate that U(VI) (as the soluble uranyl ion) is readily reduced by green rust to U(IV) in
the form of relatively insoluble UC>2 nanoparticles, suggesting that the presence of green rusts in
the subsurface may have significant effects on the mobility of uranium, particularly under iron-
reducing conditions. Lee et al. (2010) found that biogenic UO2 (uraninite) nanocrystals may be
formed as a product of a microbial reduction process in uranium-rich environments. These
results will extend the limited knowledge of microbial uraniferous mineralization and may
provide new insights into the fate of aqueous uranium complexes.
6.2.2.2 Role of Secondary Minerals
The oxidation of iron sulfides in the host rock results in the release of iron, sulfate, acidity, and
metals to solution. High aluminum and silica concentrations are also commonly encountered in
mine effluents and are the result of weathering of aluminosilicate minerals at low pH. Oxidation
and hydrolysis reactions can subsequently lead to the precipitation of a wide array of hydroxide,
sulfate, and/or hydroxysulfate minerals, depending on geochemical and biogeochemical
conditions (Nordstrom and Alpers 1999). These secondary minerals play important roles in
attenuating contaminants in the ground water.
Secondary precipitates can remove contaminants from affected waters through adsorption and/or
coprecipitation reactions. The extent to which dissolved contaminants will sorb onto secondary
precipitates as outer sphere or inner sphere complexes will vary as a function of the contaminant
Draft Technical Report 105 Revised Draft - November 26, 2012
-------
species, the secondary precipitate, pH, particle size and surface area, and the presence of other
sorbing species that may compete for adsorption sites.
Inorganic contaminants may be removed from solution due to precipitation of an insoluble phase
in which the contaminant represents a major or minor component within the solid. Examples of
secondary precipitates that form in impacted sites include oxyhydroxides [e.g., FeOOH(s)],
hydroxysulfates [e.g., Fe8O8(OH)6(SO4)(s)], sulfates [e.g., PbSO4(s)], and sulfides [e.g., ZnS(s)].
For each of these minerals, there will be a limited compositional range of ground water
chemistry over which precipitation could occur, and formation of these precipitates may compete
with other removal processes, such as adsorption.
The potential for contaminant precipitation can be estimated by evaluating the saturation state of
the ground water with respect to possible precipitate phases using a saturation-state modeling
approach. To evaluate whether ground water is oversaturated, undersaturated, or at equilibrium
with a particular phase, computer geochemical speciation models are of practical use. As an
example, consider the solubility expression for lead sulfate (anglesite). The mass-action
expression that applies to the equilibrium is:
PbSO4(s) = Pb2+ + SO42"
a , a ,
Pt>-+ SOT" ,,,
K = - — = 10
A natural water may or may not be at saturation with respect to anglesite, depending on whether
the phase is actually present, available surface area, residence time of water, and kinetic factors
that may impede dissolution and/or precipitation. If equilibrium is assumed between water and
anglesite, then the ion activity product, Q, should be the same as the equilibrium constant, Kr:
Q=a , a , =K =10 7i
Pi-' S0|
where the activity, a, of PbSO4(s) is taken to be 1 . Because ion activity products may vary by
orders of magnitude, it is often more convenient to take the logarithm of the ratio; that is, to
compute the saturation index, ,57:
where SI = 0 at equilibrium. If the water is oversaturated in a particular phase, then the ,57 is
positive, and there is a thermodynamic driving force for precipitation to occur. If the water is
undersaturated, then the SI is negative, and the mineral, if present, will tend to dissolve:
SI > 0 if oversaturated
and
SI < 0 if undersaturated
Draft Technical Report 106 Revised Draft - November 26, 2012
-------
As previously indicated, the ground water chemistry will dictate the stability of a precipitate.
Contaminant remobilization will occur as a result of dissolution of the precipitate phase, for
example, when log Q/Kr < 0. Precipitate dissolution may occur due to ground water acidification,
oxidation/reduction of precipitate components, dilution, or complexation of the precipitate
component(s) with dissolved species that form more stable compounds. A key point is that
attenuation processes involving inorganic contaminants are reversible (e.g., Gault et al. 2005;
Moncur et al. 2005). Metals taken up at the mineral-water interface can be released back into
solution. Geochemical modeling of mineral stability and contaminant adsorption/desorption
behavior can provide insight into contaminant remobilization potential due to future changes in
geochemical conditions. However, it must be noted that thermodynamic databases are often
incomplete, and thermodynamic constants for specific compounds may vary from database to
database. Thus, results from geochemical models must be carefully reviewed. In addition, the
method outlined above assumes equilibrium conditions and ignores rates (i.e., kinetics) of
mineral dissolution and precipitation. However, data are often lacking on the kinetics of
biogeochemical processes responsible for contaminant uptake and remobilization, especially data
that can be applied in field systems to predict the long-term behavior of contaminants.
With respect to predicting geochemical interactions at ISR facilities, several concerns raised by a
reviewer of the geochemical modeling of an ISR facility and presented in NUREG-6820 (NRC
2007) illustrate the potential impacts from these types of limitations. The reviewer noted that
since the applied model is nonkinetic, any bacterial influences from naturally occurring
Desulfovibria and Thiobacillus are eliminated from consideration. The comment also noted that
these influences may be as (or more) important to long-term stability than the addition of
reductant during restoration. In addition, the role of pyrite during both restoration and
stabilization was a concern, and the reviewer noted that a kinetic approach might result in
simulations that more closely compared with observed conditions.
6.2.2.3 Role of Biological Processes
The purpose of the stabilization phase of aquifer restoration is to establish a chemical
environment that reduces the solubility of dissolved constituents, such as uranium, arsenic, and
selenium. An important part of stabilization during aquifer restoration is metals reduction (NRC
2007). During uranium recovery, if the oxidized (more soluble) state is allowed to persist after
uranium recovery is complete, metals and other constituents such as arsenic, selenium,
molybdenum, uranium, and vanadium may continue to leach and remain at elevated levels. To
stabilize metals concentrations, the pre-operational oxidation state in the ore production zone
should be re-established to the extent possible. This may be achieved by adding an oxygen
scavenger or reducing agent, such as hydrogen sulfide (H2S), or through bioremediation (NRC
2007).
Bioremediation of uranium contamination has been under field and laboratory investigation since
originally proposed in the early 1990s, with most of the efforts occurring in the last 10 to
12-years (see Abdessalem et al. 1999 for a listing of references on the subject). In
bioremediation, the added bacteria use organic materials or other electron donor species in the
ground water to generate electrons that are then transferred to an electron acceptor [ideally
U(VI)] as the end product of the bacterial metabolic process. For bioremediation, bacteria are
Draft Technical Report 107 Revised Draft - November 26, 2012
-------
used in combination with electron donor materials (e.g., acetate or lactate ions) to reduce U(VI)
to U(IV) in an effort to immobilize the uranium and prevent its migration away from the initial
contaminant source area into surrounding ground waters. Once uranium is chemically reduced,
it is anticipated that the uranium will precipitate in the host rock pore space or be adsorbed on the
surface of minerals in the host rock and remain sequestered within the treated area.
Laboratory studies of bioremediation illustrate that various bacteria and electron donor materials
in contact with aqueous U(VI) result in a reduction of uranium (see Long et al. 2008 and
Yabusaki et al. 2010 for extensive summaries of these studies). In addition to bioremediation
using added organic electron donor species, reduction of U(IV) with nitrate, sulfate, and ferrous
iron species added to ground water has been reported (Moon et al. 2009, Jeon et al. 2005).
Sulfate-reducing bacteria have the ability to attenuate the movement of metals through the
precipitation of sulfide minerals (e.g., Gammons et al. 2005) and by raising the pH of the water.
This process is recognized in the restoration of ISR sites and also occurs in the natural
environment (Church et al. 2007). These results have greatly increased interest in using
bioremediation techniques and led to some field applications.
Most of the field studies testing uranium contamination bioremediation have involved situations
somewhat different than those presented for ISR operations. The source of the uranium
contamination is the introduction of U(VI) into the subsurface from activities on the surface,
such as the disposal of mill tailings or uranium wastes spilled on the ground surface or disposed
of by shallow land burial methods. For these situations, U(VI) enters the vadose zone and
penetrates to the water table, where it can migrate to contaminate surrounding ground water. In
field studies of these situations, the ground water is injected with bacteria and other chemical
additives to initiate the uranium reduction and immobilization process. For an ISR site, the
uranium is initially present in the subsurface in a chemically reduced state from the mechanisms
that originally deposited the ore body. In the ISR process, the reduced uranium is oxidized by
chemicals pumped into the ore zone and later reduced by the addition of other chemicals during
the aquifer restoration phase, with the expectation that the more mobile U(VI) will precipitate
and remain in the wellfield. Bioremediation field studies for the ISR application have not been
reported on extensively.
While laboratory results for bioremediation of U(VI) contamination were very promising,
additional studies and field testing have not met with unqualified success (Charbonneau 2009).
Uranium (VI) adsorbed on minerals in the subsurface is significantly less available kinetically
for reduction by bacteria (Liu et al. 2009, Ortiz-Bernard et al. 2004). This finding may be
particularly relevant to ISR applications, since the oxidizing conditions during operations may
result in sorption of U(VI) on the host rock and subsequent resistance of the oxidized uranium to
be reduced and stay immobile in the changing chemical environment during and after ground
water restoration. Re-oxidation of U(IV) has been observed due to reaction with sulfate, nitrate,
and ferric iron during chemical reduction of these species by other bacteria (Moon et al. 2009).
For the ISR situation, the introduction of sulfate and ferric iron species to promote chemically
reducing conditions, as well as the presence of elevated nitrate levels in some near-surface
aquifer systems, may in fact work against the long-term immobilization of uranium in a restored
ISR wellfield. During the initial oxidation process, iron sulfide minerals will also be oxidized by
the lixiviant and ferric iron remaining in the wellfield ground water during restoration would
Draft Technical Report 108 Revised Draft - November 26, 2012
-------
compete with remaining U(VI) for reduction to iron sulfides again. A study by Wan et al. (2005)
showed that the microbially mediated reduction of U(VI) to U(IV) can be transient, even under
sustained reducing conditions. These authors found that uranium was reduced during the first 80
days, but after 100 to 500 days, it was re-oxidized and solubilized, even though the microbial
community capable of reducing U(VI) was sustained.
In addition, the ISR restoration process involves injection and withdrawal of chemicals into the
oxidized wellfield under a strong flow gradient. For bioremediation efforts, this relatively rapid
flushing of ground water through the wellfield would probably severely limit the effectiveness of
bioremediation by simply removing the bacteria before they have opportunity to work
effectively. For bioremediation attempts, it may be more productive to "inoculate" a restored
wellfield with the appropriate bacteria after the typical restoration efforts are stopped and the
pumping in the wellfield ceases. The longer-term monitoring periods proposed in the rulemaking
would then allow the progress of a bioremediation step to be monitored over a longer time frame,
since only the natural flow of ground waters through the wellfield would affect the bacteria
population.
The net effect of these processes may result in a slow oxidation of uranium left in a "restored"
ISR wellfield over the long-term, supporting the use of a longer post-restoration monitoring
period than has been used previously in the industry. Additional field studies involving restored
wellfields would allow the mechanisms mentioned above to be assessed to determine their long-
term effects under actual field conditions. These studies would be welcomed and may
significantly elevate confidence that a "restored" ISR wellfield will remain that way for the long-
term.
Cameco Resources is considering bio-stimulation tests to determine the viability of naturally
occurring bacteria to re-precipitate uranium and other redox sensitive species. Prior studies by
Cameco at Smith Ranch showed that a mixture of safflower oil and ethanol or cheese whey alone
resulted in rapid reductions of uranium, arsenic, and selenium in the ground water, but these
amendments caused pump plugging problems (Cameco 2012). Cameco is proposing additional
testing using other organic molecules to promote bio-stimulation. Cameco noted that studies at
various DOE sites involved ground water with substantially different geochemistry than that at
ISR facilities. In addition, the proposed Smith Ranch studies involved deeper-lying waters that
are less susceptible to variations in oxygen contamination than the near-surface waters involved
in the DOE studies.
6.3 Geochemically Based Restoration Techniques
Another component of aquifer restoration is accomplished by establishing a chemical
environment that alters the solubility of dissolved constituents, such as uranium, arsenic, and
selenium. These methods typically invoke chemical reactions in which the valence state of
elements is either oxidized to a higher valence state or reduced to a lower valence state.
During uranium recovery, if the oxidized (more soluble) state is allowed to persist after uranium
recovery is complete, metals and other constituents such as arsenic, selenium, molybdenum,
uranium, and vanadium may continue to leach and remain at elevated levels. For example, if
Draft Technical Report 109 Revised Draft - November 26, 2012
-------
arsenic concentrations in mildly oxidizing water down gradient from an ISR facility must be
lowered, then either increasing the redox potential to precipitate a less soluble arsenic oxide or
reducing the redox potential and adding sulfide to form a less soluble sulfide mineral might be
considered. Some of the issues to consider in the applied redox approach are the type and amount
of reactant, means of emplacement, reaction kinetics, unwanted byproducts, solubility of
contaminant-containing minerals, and geochemical stability of the imposed barrier environment.
As noted above, another method used to stabilize metals by the re-establishment of their pre-
operational oxidation states is to add an oxygen scavenger or reducing agent [such as hydrogen
sulfide (H2S)] or a biodegradable organic compound (such as ethanol) into the uranium
production zone during the later stages of recirculation (NRC 2007).
As described in the case studies summarized in NRC (2007), sampling at some sites after
injection indicated that although reducing conditions were apparently achieved, they were not
maintained over the longer term. For example, as a field test of ground water stabilization during
aquifer restoration, hydrogen sulfide gas was injected as a reductant into the Ruth ISR research
and development facility in Campbell County, Wyoming. After 6 weeks of hydrogen sulfide
injection, the pH dropped relatively quickly from 8.6 to 6.3, and the sulfate concentration
increased from 28 parts per million to 91 parts per million, indicating that the sulfide reductant
was being consumed (NRC 2007). Concentrations of dissolved uranium, selenium, arsenic, and
vanadium decreased by at least one order of magnitude. After 1 year of monitoring, however,
reducing conditions were not maintained, and uranium, arsenic, and radium concentrations began
to increase, suggesting that the amount of hydrogen sulfide injected was not sufficient to fully
reduce all the material oxidized during the mining phase.
Based on the available field data from aquifer restoration, NRC (2007) concluded that aquifer
restoration is complex and results can be influenced by several site-specific hydrological and
geochemical characteristics, such as pre-operational baseline water quality, lixiviant chemistry,
aquitard thickness and continuity, aquifer mineralogy, porosity, and permeability. In some cases,
such as at Bison Basin and Reno Creek, the aquifer was restored in a relatively short time. In
other cases, restoration required much more time and treatment than was initially estimated (e.g.,
the A- and C-Wellfields at the Highland ISR facility).
6.4 Monitored Natural Attenuation
MNA refers to the reliance on natural attenuation processes to achieve site-specific remediation
objectives within a reasonable time frame. Natural attenuation processes include a variety of
physical, chemical, and/or biological processes that act without human intervention to reduce the
mass or concentration of contaminants in soil and ground water. These in-situ processes include
biodegradation, dispersion, dilution, sorption, and volatilization; radioactive decay; and chemical
or biological stabilization, transformation, or destruction of contaminants (EPA 1999a).
The overall impact of natural attenuation processes at a given site can be assessed by evaluating
the rate at which contaminant concentrations are decreasing either spatially or temporally.
Guidelines included in Office of Solid Waste and Emergency Response (OSWER) Directive
9200.4-17P (EPA 1999a) and by the American Society for Testing and Materials (ASTM 1998)
have endorsed the use of site-specific attenuation rate constants for evaluating natural attenuation
Draft Technical Report 110 Revised Draft - November 26, 2012
-------
processes in ground water. The EPA directive on the use of MNA at Superfund, RCRA, and
underground storage tank sites (EPA 1999a) includes several references to the application of
attenuation rates:
Once site characterization data have been collected and a conceptual model
developed, the next step is to evaluate the potential efficacy of MNA as a remedial
alternative. This involves collection of site-specific data sufficient to estimate with
an acceptable level of confidence both the rate of attenuation processes and the
anticipated time required to achieve remediation objectives. At a minimum, the
monitoring program should be sufficient to enable a determination of the rate(s)
of attenuation and how that rate is changing with time.
Site characterization (and monitoring) data are typically used for estimating attenuation rates.
The ASTM Standard Guide for Remediation of Ground Water by Natural Attenuation at
Petroleum Release Sites (ASTM 1998) also identifies site-specific attenuation rates as a
secondary line of evidence of the occurrence and rate of natural attenuation.
The 1999 OSWER Directive also provides some general guidelines for use of MNA as a
remedial approach for inorganic contaminants (EPA 1999a). The key policy concerns are that the
specific mechanisms responsible for attenuation of inorganic contaminants should be known at a
particular site, and the stability of the process should be evaluated and shown to be irreversible.
The specific policy language is as follows:
MNA may, under certain conditions (e.g., through sorption or oxidation-reduction
reactions), effectively reduce the dissolved concentrations and/or toxic forms of
inorganic contaminants in groundwater and soil. Both metals and non-metals
(including radionuclides) may be attenuated by sorption reactions such as
precipitation, adsorption on the surfaces of soil minerals, absorption into the
matrix of soil minerals, or partitioning into organic matter. Oxidation-reduction
(redox) reactions can transform the valence states of some inorganic
contaminants to less soluble and thus less mobile forms (e.g., hexavalent uranium
to tetravalent uranium) and/or to less toxic forms (e.g., hexavalent chromium to
trivalent chromium). Sorption and redox reactions are the dominant mechanisms
responsible for the reduction of mobility, toxicity, or bioavailability of inorganic
contaminants. It is necessary to know what specific mechanism (type of sorption
or redox reaction) is responsible for the attenuation of inorganics so that the
stability of the mechanism can be evaluated. For example, precipitation reactions
and absorption into a soil's solid structure (e.g., cesium into specific clay
minerals) are generally stable, whereas surface adsorption (e.g., uranium on
iron-oxide minerals) and organic partitioning (complexation reactions) are more
reversible. Complexation of metals or radionuclides with carrier (chelating)
agents (e.g., trivalent chromium with EDTA) may increase their concentrations in
water and thus enhance their mobility. Changes in a contaminant's concentration,
pH, redox potential, and chemical speciation may reduce a contaminant's
stability at a site and release it into the environment. Determining the existence,
and demonstrating the irreversibility, of these mechanisms is important to show
that a MNA remedy is sufficiently protective.
Draft Technical Report 111 Revised Draft - November 26, 2012
-------
6.4.1 Tiered Approach to Assessing Suitability of Monitored Natural Attenuation
EPA's Office of Research and Development has prepared a technical resource document for the
application of MNA to inorganic contaminants in ground water (Reisinger et al. 2005; EPA
2007a and 2007b). The technical resource document presents a four-tiered assessment of MNA
as a viable response action for selected metal, metalloid, and radionuclide contaminants
encountered in ground water at a particular location. Components of the approach common to
each tier include (1) demonstrating contaminant sequestration mechanisms, (2) estimating
attenuation rates, (3) estimating attenuation capacity of aquifer solids, and (4) evaluating
potential reversibility issues. EPA expects that users of this document will include EPA and state
cleanup program managers and their contractors, especially those individuals responsible for
evaluating alternative cleanup methods for a given site or facility. A decision-making approach is
provided for evaluating MNA as a possible response action for contaminated ground water.
Emphasis is placed on developing a more complete understanding of the site through
development of a conceptual site model that includes an understanding of the attenuation
mechanisms, the geochemical conditions governing these mechanisms, and indicators that can be
used to monitor attenuation progress (EPA 2007a).
EPA judges this tiered decision-making approach to be an appropriate and cost-effective way to
screen out sites unsuitable for MNA, while collecting the most relevant data at sites that might be
amenable to this approach. Conceptually, a tiered assessment of MNA seeks to progressively
reduce site uncertainty as MNA-specific data are collected. MNA for inorganics and
radionuclides is most effectively implemented through four tiers that require progressively more
information on which to assess the reasonableness of MNA:
• Tier I. The plume is not threatening public health, is stable, and some direct evidence of
contaminant attenuation exists.
• Tier II. The attenuation capacity of the site exceeds the estimated mass of contaminant at
the site.
• Tier III. There is strong evidence that attenuation mechanism(s) will prevail over long
periods of time.
• Tier IV. A record of decision, including a long-term monitoring plan and other site
closure considerations, is developed.
6.4.2 First-Order Attenuation Rate Determination
First-order attenuation rate constant calculations are an important consideration for evaluating
natural attenuation processes at ground water contamination sites. Specific applications
identified in EPA guidelines (EPA 1999a) include use in characterization of plume trends
(shrinking, expanding, or showing relatively little change), as well as estimation of the time
required to achieve remediation goals. As described by Newell et al. (2002), the use of the
attenuation rate data for these purposes is complicated, as different types of first-order rate
constants represent very different attenuation processes:
Draft Technical Report 112 Revised Draft - November 26, 2012
-------
Concentration versus time rate constants are used for estimating how quickly remediation goals
will be met at a site. In units of inverse time (e.g., per day), they are derived as the slope of the
natural log concentration versus time curve measured at a selected monitoring location.
Concentration versus distance bulk attenuation rate constants are used for estimating whether a
plume is expanding, showing relatively little change, or shrinking because of the combined
effects of dispersion, biodegradation, and other attenuation processes. The attenuation rate
constant, in units of inverse time (e.g., per day), is derived by plotting the natural log of the
concentration versus distance and (if determined to match a first-order pattern) calculating the
rate as the product of the slope of the transformed data plot and the ground water seepage
velocity contaminant transport versus transport of a tracer, or more commonly, calibration of a
solute transport model to field data.
To interpret the past behavior of plumes, and to predict their future behavior, it is necessary to
describe the behavior of the plume in both space and time. Therefore, the collection of long-term
monitoring data from wells that are distributed throughout the plume is important. Concentration
versus time rate constants describe the behavior of the plume at one point in space, while
concentration versus distance rate constants describe the behavior of the entire plume at one
point in time. Under appropriate conditions, each of these constants can assist in site-specific
evaluation and quantification of natural attenuation processes. Each of these terms is identified as
an "attenuation rate." Because the rate constants differ in their purpose and relevance, it is
important to understand their proper application, as summarized below.
Concentration versus Time Rate Constants: A rate constant derived from a concentration versus
time (C vs. T) plot at a single monitoring location provides information regarding the potential
plume longevity at that location, but that information cannot be used to evaluate the distribution
of contaminant mass within the ground water system. The C vs. T rate constant at a location
within the source zone represents the persistence in source strength over time and can be used to
estimate the time required to reach a remediation goal at that particular location. To adequately
assess an entire plume, monitoring wells must be available that adequately delineate the entire
plume, and an adequate record of monitoring data must be available to calculate a C vs. T plot
for each well. At most sites, the rate of attenuation in the source area is slower than the rate of
attenuation of materials in ground water, and plumes tend to shrink back towards the source over
time. In this circumstance, the life cycle of the plume is controlled by the rate of attenuation of
the source and can be predicted by the C vs. T plots in the most contaminated wells. At some
sites, however, the rate of attenuation of the source is rapid compared to the rate of attenuation in
ground water. This pattern is most common when contaminants are readily soluble in ground
water and when contaminants are not biodegraded in ground water. In this case, the rate of
attenuation of the source as predicted by a C vs. T plot will underestimate the lifetime of the
plume. This behavior would be expected at ISR sites, following the remediation of the source.
Concentration versus Distance Rate Constants: Attenuation rate constants derived from
concentration versus distance (C vs. D) plots serve to characterize the distribution of contaminant
mass within space at a given point in time. A single C vs. D plot provides no information with
regard to the variation of dissolved contaminant mass over time and, therefore, cannot be
employed to estimate the time required for the dissolved plume concentrations to be reduced to a
specified remediation goal. This rate constant incorporates all attenuation parameters (sorption,
Draft Technical Report 113 Revised Draft - November 26, 2012
-------
dispersion, biodegradation) for dissolved constituents after they leave the source. Use of the rate
constant derived from a C vs. D plot (i.e., characterization of contaminant mass over space) to
characterize contaminant mass over time will provide erroneous results. The rate constant based
on C vs. D indicates how quickly dissolved contaminants are attenuated once they leave the
source, but provides no information on how quickly a residual source zone is being attenuated.
Most sites will have some type of continuing residual source zone, even after active remediation,
making the C vs. D rate constant inappropriate for estimating plume lifetimes for most sites.
In assessing the potential for long-term stability of a restored wellfield, and the potential in the
exempted aquifer to contain any re-mobilized uranium from migrating into portions of the non-
exempted aquifer, development of a geochemical model for the site is of critical importance.
Natural attenuation mechanisms can play a role in limiting the potential for mobilized uranium,
and other contaminant species, from escaping the exempted aquifer and potentially
compromising potable water resources. In the absence of a geochemical model which
demonstrates an adequate reducing capacity in the restored wellfield and down-gradient portion
of the exempted aquifer, re-mobilization of uranium may be possible if the original mining
process removed the mechanisms that were responsible for sequestering the uranium initially.
For such a situation, restoration and monitoring may not detect the slow degradation of the
restored wellfield during only a limited period of post-restoration monitoring. The 30-year
proposed post-restoration monitoring period allows an opportunity to detect a slow degradation
and institute remediation measures through the regulatory process. If the degradation process
occurs after the 30-year period, remediation efforts and its costs would fall on the taxpayer. To
avoid that situation, demonstrating the reducing capacity of the restored exempted aquifer is of
particular importance.
Draft Technical Report 114 Revised Draft - November 26, 2012
-------
7.0 STATISTICAL ANALYSES TO COMPARE PRE- AND POST-ISR CONDITIONS
Statistical techniques and measures are frequently used to determine compliance with regulatory
requirements. Because of the inherent variations in natural conditions and media properties in
any particular location, confidence intervals appropriate to these specific statistical measures and
available databases are also employed to allow a degree of flexibility in regulatory decisions
accounting for natural variability. We have adopted this confidence measure approach in the
Part 192 rulemaking for ISR operations. The statistical techniques selected for regulatory
compliance demonstrations are dependent on the quality and quantity of field data available for
the analyses. Selection of appropriate statistical techniques must be justified by considerations of
the available data for building a defensible compliance compliance case. This chapter examines
the applicability of various statistical techniques and measures to address various issues involved
in ISR operations and regulatory decisions.
Note: The statistical discussions in this chapter include traditional usage of many
symbols like M and m, N andn, P andp, Tandt, a and ft, and S and a. The
definition of these symbols differs from section to section. The meaning of each
symbol should be clear from the context of the discussion.
The statistical methods discussed in this chapter provide tools for answering the following basic
questions of particular relevance to regulatory decisions for ISR operations:
(1) What is the baseline at this site?
(2) Has the site been restored to baseline?
(3) Are there trends that indicate the site may not stay at baseline levels?
(4) If one or more wells exceed the baseline, do trend analysis and/or modeling suggest the
site will return to baseline [or an alternate concentration limit (ACL) as specified in
40CFR264.94(b)]?
The answers to these questions should be based on data sufficient for the purpose. This chapter
presents methods for determining what is a sufficient number of samples to answer questions
such as these. The chapter also presents statistical methods for analyzing the data to reach a
conclusion.
Due to sampling variability, decisions such as these may be difficult to make without the use of
hypothesis testing. Hypothesis testing provides a framework for controlling the frequency of
decision errors. The procedures in this document favor protection of the environment and human
health. If uncertainty is large or the sampling inadequate, these methods conclude that the
sampled area does not attain the cleanup standard.
Hypothesis testing is used as a statistical tool for deciding when the ground water has reached
steady state and for comparing post-restoration conditions with predetermined restoration goals.
The statistical tests are based on measurements of baseline and post-restoration water quality
conditions at the site. These measurements include a wide variety of water quality parameters.
Draft Technical Report 115 Revised Draft - November 26, 2012
-------
Usually, the measured parameter is a concentration of an analyte in a specific well at a given
time, but other water quality properties may also be analyzed using the methods in this chapter.
Statistical tests are required to make decisions due to spatial and temporal variability in the
sample values. Two types of statistical hypothesis tests are employed: two-sample comparisons
of post-restoration and baseline data are used to determine if restoration goals have been
attained, and trend analysis is used to determine if stability is achieved. Several two-sample
statistical tests are described for comparing pre- and post-conditions in individual wells. A
heterogeneity test then is used to determine if the test results are consistent across all wells in the
unit. Once restoration goals are attained, subsequent monitoring is required to demonstrate
stability. Several statistical trend tests are used to determine if stability is maintained. The trend
tests address temporal variability when determining if restoration goals are maintained.
Although statistical analyses are used in all phases of the ISR process shown in Figure 3-1, three
phases employ formal statistical methods to characterize baseline conditions, to determine when
compliance with the baseline has been achieved, and when long-term stability has been
demonstrated.
• Phase 1 - Measure baseline ground water concentrations to within established precision
objectives and establish restoration goals based on statistical procedures that address
temporal and spatial variability.
• Phase 4 - Establish wellfield restoration. At the end of this phase, the ground water
potentiometric surface will have returned to baseline conditions, and statistical tests for
significant differences are used to verify restoration.
• Phase 5 - Conduct long-term stability monitoring. At the end of this phase, statistical
tests for trends are used to show that concentration of the monitored parameter is not
increasing (or, in some cases, decreasing) with time.
Procedures used to combine data from separate wells to determine whether the site as a whole
attains the restoration goals are discussed. Testing the samples from individual wells or groups of
wells is also discussed.
Table 7-1 shows an outline of the statistical procedures used in Phase 1, Phase 4, and Phase 5.
The first step in each phase requires estimation of the number of samples required for the task.
The number of samples are determined by site conditions and the Data Quality Objectives
(DQOs) established for each task.
Each phase has a data collection step. Data collection should be governed by appropriate quality
control guidelines. 40 CFR 146 (Subpart D) specifies construction, operation, monitoring, and
reporting requirements for injection wells (Class III). The Agency-wide EPA program
requirements for quality assurance are described in EPA 2000c. EPA 2000d provides guidance
for establishing data quality objectives (DQOs). EPA 2001 and EPA 2002d provide guidance for
developing a Quality Assurance Project Plan (QAPP). EPA 2002e and EPA 2006b provide
additional guidance for data collection activities. A detailed sampling plan should be submitted
and approved prior to beginning data collection activities. All data collected in each phase of
sampling are to be retained and submitted in standard digital format, such as a spreadsheet or
Draft Technical Report 116 Revised Draft - November 26, 2012
-------
database. The data should include well identification, sample collection date and time, analysis
date and time, well location and screening interval, units of measurement, detection limits, and
identification of outliers and nondetects. As many of the statistical tests require modifications
when there are a large number of ties, unnecessary rounding of sample values should be avoided.
Table 7-1. Outline of the Statistical Procedures used in Phases 1, 4, and 5
Phase 1
Determine Baseline
1) Determine number of
baseline samples
(Section 7.1)
2) Collect samples and
estimate baseline parameters
(Section 7.1.2)
3) Check for unexpected
trends (Section 7.7.2)
4) Define survey units and set
restoration goals
(Section 7.1)
Phase 4
Compliance with Baseline
1) Determine number of
monitoring wells and samples
per well (Sect 7.2 and 7.6)
2) Collect samples and
compare with baseline using
(a) Two-sample t-te&t,
(b) Prediction limits, or
(c) Wilcoxon Rank Sum test
(Section 7.9)
3) Check for unexpected
trends (Section 7.7.2)
Phase 5
Stability Monitoring
1) Determine number of
samples for detecting a trend
(Section 7.7.2)
2) Collect samples and test
for significant trends using
(a) Regression Mest, or
(b) Mann Kendall test
('Sect. 7.7.2 and 7.7.31
Statistical hypothesis tests are recommended for comparing post-restoration conditions with
baseline conditions and for demonstrating stability of the site after restoration. Several
parametric and nonparametric statistical tests are presented for the comparison with baseline
conditions. These tests are used in Phase 4 to determine if the restoration goals have been met.
The two-sample t-test and prediction limits (PLs) for a future mean are two parametric methods
used in the comparison with baseline conditions, assuming that both data sets are stationary. The
1 9
nonparametric Wilcoxon Rank Sum (WRS) test is also used to compare post-restoration well
conditions with baseline values. The two-sample t-tesi and the WRS test are recommended for
comparing baseline and post-remedial wells (EPA 2006a). The RCRA Unified Guidance (EPA
2009) recommends the PL method.
Linear regression and the nonparametric Mann-Kendall trend tests are recommended for trend
detection in EPA 2006a and EPA 2009. The linear regression trend test relies on a variety of
assumptions (e.g., normality) that need to be verified. The Mann-Kendall trend tests may be used
with any series of four or more independent samples to test for trends in well parameters. The
trend tests are used in Phase 1 to check for unexpected trends in baseline samples, and in Phase 5
to establish long-term stability. As an example of unexpected variations, sampling before
operations in the ore zone and up gradient along with continued sampling up gradient of the
' This test is known also as the Mann-Whitney or Wilcoxon-Mann-Whitney test.
Draft Technical Report
117
Revised Draft - November 26, 2012
-------
production zone during ISR operations and during restoration can supply important information
about seasonal variations in the ground water chemistry for shallow aquifer ore deposits. The
Mann-Kendall trend test has been applied in ground water monitoring at RCRA sites (HydroGeo
Logic 2005).
Several EPA sources were used as the bases for the statistical tests. Although these sources do
not recommend procedures for ISR sites in particular, the sources are either general in nature or
address related issues. The sources include guidance for applying the DQOs at remediated
Comprehensive Environmental Response Compensation and Liability Act (CERCLA) sites
(EPA 2002a), guidance for conducting the statistical tests in the Multi-Agency Radiation Survey
and Site Investigation Manual (MARSSIM) (EPA 2000a), guidance for statistical analysis of
ground water monitoring data at RCRA facilities (EPA 1989, EPA 2009), and general guidance
for the application of nonparametric statistical tests found in Data Quality Assessment: Statistical
Methods for Practitioners, EPA QA/G-9S (EPA 2006a). Many of the procedures for conducting
the tests cited here were adapted from the EPA QA/G-9S document and are considered in more
detail in Attachment D. We are not mandating the use of specific statistical methods in the
Part 192 rulemaking, but we must emphasize that whatever statistical methods are selected by
the operator for preparing a compliance case must be rigorously justified on the basis of the
quantity and quality of the database collected. It is the responsibility of the appropriate
regulatory authority to make the judgement on the reliability and defensibility of the statistical
analyses presented by the ISR operator for regulatory approval.
In summary, the statistical approaches for each phase are:
Phase 1 Baseline Sampling
• Estimate required number of baseline samples (Section 7.1)
• Adjust measured data for seasonality if required (Section 7.7.1 and Attachment D,
Section D.I)
• Use regression trend test or Mann-Kendall test to check for unexpected trends
(Section 7.7.2 and Attachment D, Sections D.2 and D.3)
Phase 4 Establish Compliance with Baseline
• Estimate required number of monitoring wells (Section 7.2)
• Adjust measured individual well data for seasonality, if required (Section 7.7.1 and
Attachment D, Section D. 1).
• Use the two-sample f-test (Section 7.9.1.1), PLs (Section 7.9.1.2) or the WRS test
(Section 7.9.2.1) to compare baseline to post-restoration conditions for each well or for
pooled wells. (Attachment D, Section D.4)
• For multiple wells, first test wells for homogeneity. If the hypothesis of homogeneity
across all wells is accepted, then test to confirm compliance of all wells with restoration
goals. (Section 7.9.2.2 and Attachment D, Section D.5)
Draft Technical Report 118 Revised Draft - November 26, 2012
-------
• If steady-state data are from different wells than the baseline data and trends are not
detected, the two-sample t-tesi or the WRS test may be used to compare baseline to
steady-state measurements for statistical differences for the pooled data of all wells
combined, which are treated as a single well. (Sections 7.9.1 and 7.9.2, and Attachment
D, SectionD.4)
Phase 5 Long-term Stability Monitoring
• Determine sampling frequency and duration (Section 7.7.2)
• Adjust measured data for each well for seasonality if required (Section 7.7.1 and
Attachment D, Section D.I)
• Use Mann-Kendall or regression trend test to test for trends for each well or in the pooled
wells (Section 7.7.2 and Attachment D, Sections D.2 and D.3)
• If trend is detected, use linear regression or Theil-Sen test to assess trend magnitude
(Section 7.7.2)
• Repeat for each well
• If the multiple wells are evaluated, use the pooled-regression trend test or Mann-Kendall
test for multiple locations (Section 7.7.2)
Post-restoration samples are expected to have a higher degree of variability and trend than found
in baseline samples. Accordingly, periodic measurements for each contaminant [see proposed 40
CFR 192.52] and other species needed for supporting assessments, such as geochemical
modeling] should be taken from each well over the initial post-restoration period. It is anticipated
that the sampling will be quarterly, with four samples per year at each well. Quarterly sampling
permits analysis of the data for seasonal variations to determine if variations in measurements
reflect normal seasonal variability and not an increase in contaminants. Quarterly sampling is
typical practice in the industry and this assumption was used in performing the illustrative
calculations presented in this document. Analyses of quarterly sampling and assumptions about
natural variability (Tables 7-19 to 7-21, Section 7.7.2.2) suggest that quarterly sampling to reach
the required level of confidence about the presence or absence of trends may require very long
periods for post-restoration monitoring. More frequent sampling would reduce the monitoring
period in these situations; however, more frequent sampling must still assure that the individual
samples are independent, as discussed further below.
7.1 Determine Baseline Characteristics
The baseline characterization provides the frame of reference against which post-operational
ground water remediation is judged. The goal of baseline sampling is to establish a zone-specific
statistical distribution of baseline concentrations for key constituents and other hydrogeological
parameters. Current guidelines require these distributions be based on independent and
representative water samples collected from zones in which baseline wells are located by a
statistically valid sampling design. It is emphasized here that baseline sampling refers to
sampling within an ore zone that will be mined and subsequently restored. It should not be
confused with background sampling, which defines water quality over a broader area and
Draft Technical Report 119 Revised Draft - November 26, 2012
-------
includes, for example, up gradient and down gradient wells or sampling points located above or
below the mined zone.
Leachable uranium deposits are found in sandstones that have been deposited in basins, along
mountain fronts, and in near-shore marine and deltaic environments. The rock sequence
containing the ore bodies can be composed of a number of water-bearing units separated by
confining units. The water-bearing unit containing an ore body is separated (at least locally) from
other water-bearing units above and below.
The ore zones follow the general trend of drainage channels as shown in Figure 7-1. The shapes
vary from almost linear to serpentine, reflecting the local draining. Individual ore bodies in
sandstone lenses rarely exceed a few hundred meters in length. These are typically "roll-front"
deposits, with reducing conditions on the down gradient side and oxidizing conditions on the up
gradient side (see Figure 4-5 in Section 4.1.1). Uranium is continuously dissolved by oxygenated
ground water and displaced further down gradient. As the uranium comes in contact with the
reducing conditions down gradient, an economically recoverable deposit of uranium may
eventually be formed. The term "roll front" is used because over time, the redox interface rolls
down gradient as more oxygen is transported into the aquifer. The inner contact of ore and
altered sandstone is generally sharp, whereas the uranium concentration on the reduced side of
the interface is gradational.
Although leaching mobilizes and removes a portion of the uranium in the ore body, a large
fraction of the uranium remains within the host rock after economically feasible extraction is
completed. The concentration of uranium appearing in water samples from the baseline and post-
restoration monitoring wells is strongly influenced by redox conditions in the ore body at the
time of sampling. Other analytes have unique ground water chemistries also, with many inter-
relationships between the analyte concentrations, host rock, hydrological parameters and redox
conditions. It is possible that wells exhibiting higher radionuclide concentrations (uranium and
radium particularly) in the pre-operational time frame may show significantly lower
concentrations after production and, conversely, lower concentrations in some wells may become
higher due to the non-homogeneous characteristics within the wellfield (local effective
porosities, flow paths, changed geochemical conditions in the pre- and post-operational time
frames). A major purpose of the baseline characterization is to determine the presence and
contaminant concentrations of "hot spots," so that these data can be used in developing
restoration goals. A rigorous baseline characterization would avoid problems later during
restoration that might occur if the range of baseline contaminant concentrations were too low
because the "hot spots" were not sampled initially.
The current state-of-the-art for analyzing spatial variability is the use of geostatistical methods.
Although a full geostatistical analysis is not required, these methods provide a way to visualize
the sample data using 2- and 3-D graphical representations of the entire ore zone. Integration of
additional sample data from areas outside the ore zone and in over- and under-lying strata
provides insight into the differences that may exist between baseline characteristics within the
ore zone and baseline characteristics in surrounding regions. For those operators with an
understanding of geostatistical software and analytical procedures, these procedures may provide
Draft Technical Report 120 Revised Draft - November 26, 2012
-------
better insight into baseline characteristics and the inter-relationships of the analytes and
hydrogeological parameters under baseline conditions.
Baseline monitoring wells should be spaced sufficiently far apart that the ground water chemistry
measurements in a well are not influenced by adjacent wells. NRC's current expectations
regarding baseline sampling in the ore zone are set out in their Standard Review Plan for In Situ
Leach Uranium Extraction License Applications (NRC 2003, p. 5-39), which specifies that one
well per acre should be sampled. A single sample from each baseline monitoring well is
insufficient to determine whether water-quality parameters are stable and representative of the
ground water at the sample location. Baseline chemistry is established based on a statistical
analysis of ground water data from a sufficiently large set of wells sampled over a period of time.
The RCRA requirements (40 CFR Part 264 Subpart F) for the frequency of sampling are
applicable for this purpose: A sequence of at least four samples, taken at an interval that assures
that an independent sample is obtained, by reference to the uppermost aquifer's effective
porosity, hydraulic conductivity, and hydraulic gradient, and the fate and transport characteristics
of the potential contaminants. Calculations of the length of time or distance required to ensure
samples will be independent should be provided in support of the well locations and sampling
dates proposed in the baseline sampling plan.
The baseline data provide the technical basis for establishing restoration goals for the post-
restoration monitoring phase. As such, it is critical that the baseline data represent the natural
variability of each analyte, unbiased by variability resulting from residual effects of drilling,
construction and development. Under some conditions, residual impacts from drilling can
dominate the concentrations of some ground water constituents (particularly trace metals) in the
vicinity of the well screen for months (if not years).
The re-equilibration time of baseline wells should be confirmed prior to sampling. The statistical
tests for trends found in Section 7.3.4 may be applied to demonstrate steady-state geochemical
conditions at baseline wells. Additional guidance on best practices for baseline ground water
sampling is found in Puls and Barcelona (1989) and Yeskis and Zavala (2002). These documents
give emphasis to the following technical aspects of baseline sampling:
• Documenting the volume of water purged before sample collection and field
parameter data measured during purging (e.g., pH, Eh, conductivity, turbidity) to
provide a basis for assessing whether the ground water sample is representative of
predrilling conditions.
• Collecting additional water quality samples during purging that may provide
additional insights on well performance issues. To this purpose, a time series of
water quality samples during purging may be analyzed for major ions, trace metals
and nonmetals, and total organic carbon. These data would then be evaluated for
trends that might indicate that residual drilling or construction products remain, that
mixing of ground waters from different hydrologic zones has occurred, or
disequilibrium is evident in formation mineralogy.
Draft Technical Report 121 Revised Draft - November 26, 2012
-------
• Other approaches for evaluating representativeness of baseline well data include
plots of redox-couple data on phase diagrams and use of geochemical modeling to
determine the extent to which measured baseline water quality parameters are in
equilibrium or disequilibrium with mineral phases known to be present in the
formation.
An adaptive approach to baseline characterization is recommended that builds on information
previously collected. For example:
• A need for additional background well locations may be determined by the level of
uncertainty in the range and spatial variability of ground water constituents;
• A need for additional data from a particular well (or the need to resample or replace
a well) may be determined by the consistency of the sample data with
concentrations predicted from nearby wells; and
• A need to continue sampling an individual well may be determined by testing for
trends in the data indicating the extent to which the well has recovered from
drilling and construction activities.
After samples have been collected and analyzed by the laboratory, the data should be inspected
for unusual values that may unduly influence estimates of the baseline conditions. In many cases,
laboratory results may indicate that the concentration is below the detection limit for the
analytical method applied. These samples are commonly called "nondetects." Nondetects should
be included in the calculation of baseline concentration distribution and in post-restoration
comparisons with the baseline. If the nonparameteric methods recommended in this chapter are
applied, nondetects may be included with no modifications. For parametric procedures, a value
equal to one-half of the level of detection may be used for the nondetect samples.
7.1.1 Design for Baseline Sampling
Baseline conditions are characterized by the distribution of baseline samples collected in the ore
zone. It is important to ensure that the baseline sampling program provides samples that are
representative of ore zone conditions. The location of the baseline wells is based on a statistically
valid sampling design developed following the DQO process, as described in Section 7.1.3. A
random selection of wells from a systematic grid is one example of a statistically valid approach
for locating baseline wells.
The design and implementation of a baseline characterization program will be driven by a variety
of site-specific factors. The design for the baseline sampling program should include the number,
location, and density of baseline monitoring wells and the timing of the samples. The density of
monitoring locations will depend on the spatial variability of the analytes to be monitored:
greater variability requires higher monitoring density. In this regard, each ISR site will have
unique characteristics; hence, a flexible approach to designing the baseline sampling program is
required. Although randomness requirements will serve to avoid bias, it is important to look for
Draft Technical Report 122 Revised Draft - November 26, 2012
-------
other complicating geological factors that do not conform to randomness. Complicating factors
include, but are not limited to:
• Intersecting or adjoining deposits near mine leases
• Ground water contamination in adjacent abandoned mine shafts and tunnels
• Dewatering effects of old mine workings in or near a proposed ISR operation
• Limited knowledge about site mineralogy, particularly related to trace metals
• Changes occurring in the ground water environment unrelated to the mine itself
As sampling progresses, the baseline sampling plan may be modified to adapt to unexpected
spatial and temporal variability by increasing either the density of monitoring locations or the
period of time each location is sampled.
7.1.2 Selection of Baseline Monitoring Wells
The methods discussed in Sections 7.1.3 and 7.1.4 will be used to determine the required number
of baseline monitoring wells. It is also necessary to specify the locations of these wells. There are
three cases to consider when selecting the baseline monitoring wells:
(1) All baseline monitoring wells will be selected from pre-existing wells;
(2) Some or all of the pre-existing wells in the wellfield area will serve as baseline
monitoring wells, but additional monitoring wells are required; and
(3) No pre-existing wells will be used as baseline monitoring wells, and new locations
within the wellfield area are to be selected.
If the wellfield has a simple geometry, then a rectangular grid with the appropriate number of
sampling locations is designed to cover the wellfield as uniformly as is technically feasible. The
grid method is easy to apply, but requires modification when applied in Cases 1 and 2, or in
Case 3 with a complex wellfield geometry.
In Cases 1 and 2, some of the wells already in place are to be selected as baseline monitoring
wells, and prior information is available for these wells. This situation has both advantages and
disadvantages. One advantage is that data from these wells may be used to estimate the
variability of each analyte, which is used in Section 7.1.3 to determine the number of baseline
monitoring wells. However, a serious disadvantage is that availability of prior information allows
for the possibility of selection bias in characterizing baseline conditions by the purposeful
selection of pre-existing wells, which tend to exhibit high (or low) readings for certain
characteristics. In these cases, use of a randomized or systematic grid procedure for selecting
pre-existing wells to serve as baseline monitoring wells is recommended to avoid the possibility
of selection bias.
In Case 3, the geometry of the site is of greatest concern. For wellfields with simple geometries,
a systematic grid is recommended. Creation of a systematic grid may be more difficult for
wellfields with complex geometry of the type shown for Christensen Ranch ISR Mine Unit 6 in
Figure 7.1. The collection of narrow and tortuous wellfield subareas in this unit make it difficult
Draft Technical Report 123 Revised Draft - November 26, 2012
-------
to design a grid of representative sampling locations covering the wellfield. The following
alternative designs may be useful in certain cases.
Random Design for Case 1
In this design, all baseline monitoring wells will be selected randomly from a list of TV qualified
pre-existing wells located in the wellfield area. In this design, each pre-existing well (j=l,...,N) is
assigned a random number (xj) between 0 and 1. The list is then sorted in ascending order using
the random numbers x\,.. .,XN as the sorting variable. If «o baseline monitoring wells are required,
then the first «o pre-existing wells in the list are selected as baseline monitoring wells. The main
advantage of the random design is that it is easy to implement. The main disadvantage is that
there may be large "gaps" between monitoring wells in some regions of the wellfield due to the
random selection process, while in other areas the monitoring wells may be clustered too close
together to provide independent samples. Use of a systematic sampling grid design minimizes
these disadvantages.
Systematic Design for all Cases
The following approach uses a systematic sampling grid for locating baseline monitoring wells.
The grid sampling approach may be useful for all three cases described above. A systematic
sampling grid is applied over a broad area surrounding the wellfield, but only grid points falling
within the wellfield outline are used as baseline monitoring locations. Grid points which do not
fall within the wellfield outline are ignored. In this design, a rectangular area is defined enclosing
the entire outline of the wellfield and excursion monitoring wells.13 The rectangle is filled with a
rectangular sampling grid of sufficient density that the wellfield area will contain the required
number of baseline monitoring well locations with high probability.
One baseline monitoring well is selected for each grid point lying within the wellfield outline. In
Cases 1 and 2, the pre-existing well nearest to each grid point is selected as a representative
baseline monitoring well for that grid point, but only if the pre-existing well is no further than a
specified maximum distance (see below) from the grid point, and if the pre-existing well is
qualified by meeting all baseline monitoring well requirements for construction, screening depth
and sampling procedure. A pre-existing well can only represent one sampling grid location. If a
grid point has several nearby pre-existing wells, only the one nearest to the grid point is used. In
this situation, the remaining pre-existing wells are not used as baseline monitoring wells. (If the
density or spatial distribution of pre-existing wells in Case 1 is not sufficient to assign a qualified
pre-existing well to each grid point, then Case 2 applies.)
In Case 2, the procedure for Case 1 is applied until no more qualified pre-existing wells are
available for assignment. All remaining grid points with no qualified pre-existing well assigned
13 The inclusion of surrounding excursion monitoring wells enlarges the enclosing rectangle. This step is
necessary to ensure that narrow ore zones near the edge of the rectangle will have the same chance of selection as
interior zones. If the outer edge of the ore zone is used as the boundary of the rectangle, the grid of sampling points
may always miss ore zones along the edge. No baseline monitoring wells will be located outside of the ore zones.
The orientation of the grid may be fixed by requiring that the smallest enclosing rectangle is used. This would
further reduce the possibility of selection bias.
Draft Technical Report 124 Revised Draft - November 26, 2012
-------
will require new baseline monitoring wells be constructed at or near each grid point. In Case 3,
no pre-existing wells are available, and a new baseline monitoring well is located at or near each
grid point. If the wellfield is similar in shape and size to the enclosing rectangle, then Case 3
reduces to the same design problem addressed in Sections 7.1.3 and 7.1.4.
Application to Christensen Mine Unit 6:
The Christensen Ranch Mine Unit 6 wellfield area (the highlighted area in Figure 7-1) and the
surrounding excursion monitoring wells are enclosed by a rectangle of approximate size L =
1,300 yards by W= 1,100 yards, covering an area of approximately 295 acres. The wellfield area
itself is complex in shape and encompasses approximately AO= 42 acres ~ 200,000 sq. yd.
(Table 7.4). This is approximately 14.2% of the area of the enclosing rectangle. In this example,
we will use the NRC rule of thumb of 1 baseline monitoring well per acre of wellfield, so at least
no = 42 wells are required within the wellfield. In Case 1, some or all of these 42 wells would be
pre-existing wells.
A sampling grid is constructed over the entire rectangular area using a rectangular grid. The
sampling grid is an array of rectangles with shape proportional to the enclosing rectangle. The
required number of grid rectangles (m) is given by:
where p = A(/LW= 0.142, q = 1-p, and za is the lOO-(l-a) percentage point of the standard
normal distribution.
This formula is based on the assumption that the actual number of grid points inside the wellfield
follows a binomial distribution with a relatively large number of grid points and probability p.
Due to the tortuous shape of the wellfield area, there is a chance that the wellfield may not
actually contain the expected number of grid points within its boundaries. The parameter a
determines the probability that the constructed grid will have less than no grid points inside the
wellfield. For a = 0.05 use za = 1.645. When a = 0.50, za = 0 and m = no/p. Using this value of
m, the expected number of grid points falling inside the wellfield will be 42, but the actual
number has approximately a 50% chance of being smaller than 42.
For the Christensen MU6 unit, m = 373 = (19.3)(19.3). Rounding to next highest integers, a
20 x 20 grid with 400 rectangular grid elements of size 65 yards by 55 yards is required for 95%
assurance that the wellfield area will contain at least 42 grid points for baseline monitoring well
locations. A grid sampling point is located at the center of each grid rectangle. With these
dimensions, the greatest distance between grid sampling points is approximately 85 yards. This
value would be used as the maximum distance that a pre-existing well can be from its assigned
sampling grid point in Cases 1 and 2.
Draft Technical Report 125 Revised Draft - November 26, 2012
-------
7.1.3 Determining the Number of Baseline Samples
The selection and location of the baseline sampling wells was discussed in the previous section.
This section provides a statistical approach for determining the required number of baseline
samples. A sufficient number of wells should be sampled to ensure that the baseline is
adequately characterized. In this document, adequacy of the baseline characterization is
measured by the precision of the estimates of the baseline parameter values.
The goal of the baseline sampling design is to estimate the mean baseline concentrations of all
analytes of concern to within a specified level of precision. The required level of precision is
obtained by having a sufficient number of samples to reduce the relative standard error of the
mean to within acceptable levels. If there is a large degree of heterogeneity in the baseline
concentrations in different sections of the production unit, it may not be possible to reduce the
relative error of the mean to the specified level. In that case, it may be necessary to divide the
unit into several survey units comprised of contiguous areas with similar baseline characteristics
for the analytes of greatest concern. In this case, the decisions on compliance with baseline levels
would be developed separately for each survey unit.
Relative Standard Error of the Mean
The population coefficient of variation cv = S/Mis useful because the population standard
deviation S must always be considered in the context of the population meanM. The value of
the coefficient of variation is independent of the dimensions of the measurement and is a
dimensionless number. By convention, the coefficient of variation is expressed as a percentage,
where cv = 100% is equivalent to cv = 1.0. For comparison between analytes with different units
or widely different means, the coefficient of variation provides a standardized measure of
variability of the populations.
The same advantages apply to the standard error of estimation of the mean SE^, which is equal
to S/^jN. The coefficient of variation of an estimated mean, denoted here by COVu is the ratio
of the standard error of the mean to the mean itself CO VM = SEu/M. The coefficient of variation
of the mean is often expressed as a percentage. When expressed as a percentage, the term
"relative standard error of the mean" will be used, denoted by the symbol R$EM. Thus, R&EM =
100% is equivalent to COVu = 1-0. For comparison between analytes with different units or
widely different means, the RSEu provides a standardized, unit-free non-negative numerical
value for comparing the precision with which the mean concentrations of multiple analytes have
been measured.
The value ofRSEu depends on both the standard deviation of the population S and the number of
samples N. The standard deviation is a measure of variability, and the variability in the samples
depends on many factors including temporal and spatial variations in properties of the ore zone
environment, well sampling methods, and laboratory measurement error. Many of these sources
of variability are beyond the control of the site operator. As variability depends on site-specific
characteristics, specific baseline sample sizes for a production unit cannot be determined in
advance without sufficient prior data to provide at least a rough estimate of the standard
deviation. Some potential constituents, particularly radionuclides and other matrix characteristics
Draft Technical Report 126 Revised Draft - November 26, 2012
-------
that affect the rate of extraction, may already have been measured during site exploration and
development and provide a reasonable sample size for estimating the standard deviation. Other
constituents may already have been monitored at nearby sites or sites with similar geochemical
characteristics. Others may have no historical background databases and require a round of
preliminary baseline sampling before the required sample size can be determined.
Table 7-2 contains summary statistics for the population coefficient of variation cv collected at
nine sites for 35 analytes.14 The average coefficient of variation ranges from 4% to over 250% in
this table. The upper part of this table shows the ranked statistics for analytes other than the trace
metals, which are shown in the lower part of the table. Analytes other than trace metals with the
highest coefficient of variation include MV, COs"2, U, NH4+, and Ra-226. Of this list, only
COs"2, U, and Ra-226 have a large fraction of measurements above the detection limit. Trace
metals with the highest coefficient of variation include aluminum, iron, selenium, and
manganese. Among these, only iron has a large fraction of measurements over the detection
limit.
Although many of the sources of variability are beyond the control of the site operator,
increasing the number of baseline samples provides a way to control the RSEu for all analytes of
concern to within acceptable levels. Table 7-3 shows an example of the RSEM approach. This
table provides the information required to control the degree of precision obtained for mean
baseline concentrations of 35 analytes at Christensen Ranch Mine Unit 6 (COGEMA 1996).
Baseline water quality sampling was conducted according to the requirements approved in the
Christensen Ranch Amendment Application. The application required four rounds of sampling.
The sampling was conducted over a short, 6-week period in the summer of 1996. Sampling
events were spaced 2 weeks apart; 2 of the 4 samples from each baseline well were analyzed for
a full suite of chemical parameters, while the remaining 2 samples were analyzed for a shorter
list of constituents. The 4 rounds of analytical results from the 42 baseline wells were combined
to establish baseline water quality. Table 7-2 provides a summary of the combined analytical
results from the baseline wells. The list of 35 analytes in Table 7-3 includes the concentrations of
major ions, trace metals, and radionuclides and other aquifer properties such as pH, alkalinity,
conductivity, and total dissolved solids (TDS).
The table shows the number of samples used to compute the baseline mean and standard
deviation, and the standard error and relative standard error of the mean for each analyte. At this
site, up to 168 samples were used to characterize the baseline. The RSEu obtained for the
analytes ranges from 0.3% to 7%. The radionuclides (U and Ra-226) are among the highest
RSEM percentages; 6.8% for U and 6.6% for Ra-226. Several trace metals, including iron and
aluminum, are also among the highest percentages. The high RSEM values indicate that these
analytes were measured with the least relative precision.
14 Sources: Crow Butte MU1 (Crow Butte 2000, Table 2); Highland A (Kearney 2004, Attachment A,
Appendix 3); Highland B (Power Resources 2004, Table 1); Irigaray MU1 to MU9 (Wichers 2006, Table 1);
Christensen MU2 North (Total Resources 1993, Table 2); Christensen MU3 (Malapai 1988, Table 2); Christensen
MU4 (COGEMA 1994, Table 6); Christensen MU5 (COGEMA 1995, Table 7); and Christensen MU6 (COGEMA
1996, Table 6).
Draft Technical Report 127 Revised Draft - November 26, 2012
-------
WELLFIELD DATA PACKAGE
MINE UNIT 6
CHRISTENSEN RANCH ISL PROJECT
EXISTING TWO TRACK ROAD
POWER LIIC
ruwLK Lint.
TRUNK LINE «•——.-«
MODULE BUILDINGS O **»
PERIMETER HDNITOR WELLS #*™
INTERIOR MONITOR WELLS O •"*
DESIGNATED RESTORATION WELLS
HYDRDLOG1C WELLS O *'
WELLFICLD OUTLINE
MINE UNIT BOUNDARY
MODULE BOUNDARY
Prepared for:
WDEQ - Permit to Mine No. 478
and NRC - License No. SUA-1341
Figure 7-1. Ore Zone Outline and Well Locations at Christensen Ranch Mine Unit 6
The required number of baseline samples to achieve a precision of P% is calculated using:
N = (SIMPf=(cvIP}2
where cv and P are expressed in percentage terms. For example, in Table 7-2, uranium has an
average coefficient of variation of 105%. The required number of samples to estimate the mean
uranium concentration with a precision of P = ±10% is:
N = (cv/p)2 =(105/10)2 =110
The right side of Table 7-3 shows the number of samples required for each analyte to achieve a
targeted RSEm of P%, for values of P ranging from 1% up to 100%. At Mine Unit 6, uranium and
Draft Technical Report
128
Revised Draft - November 26, 2012
-------
Ra-226 require the largest number of samples to achieve the targeted level of precision. For
example, 75 samples are required to achieve a hypothetical target precision of P = 10% for
uranium, and 74 samples are required for Ra-226. These sample sizes are the highest required to
achieve a target precision of P = 10% for all listed analytes. If 75 or more baseline samples are
collected at the mine unit, then the RSEu for all analytes is expected to be less than the target
value of P = 10%. With at least 167 baseline samples for uranium and Ra-226 at Mine Unit 6, the
achieved RSEM for U and Ra-226 is considerably less than 10% for both radionuclides.
Table 7-4 shows the results of similar analyses conducted for nine ISR sites.15 This table
summarizes results of the analyses, with emphasis on the RSEM achieved for the mean baseline
uranium and Ra-226 concentrations at each site. The table also shows the size, number of
baseline wells, number of wells per acre, and the average number of baseline samples per well.
The right side of this table contains the actual sample size and achieved precision for uranium
and Ra-226. This table also shows the sample sizes that would be required for a hypothetical
targeted precision of ±10%. The highlighting in this table indicates sites meeting a hypothetical
targeted precision goal of ±10%.
Table 7-5 contains the summary information for uranium and Ra-226 found in the bottom two
rows of Table 7-3 for nine ISR sites. The table shows the number of samples required for the
RSEM for uranium and Ra-226 to be less than ±P%. The table also shows the actual number of
baseline samples and the achieved precision for uranium and Ra-226 at each site. Figure 7-2
contains a scatter plot of the maximum RSEM for uranium and Ra-226 mean concentrations
versus the baseline sample size at the nine production units listed in Table 7-5. A baseline
sample size of approximately 80 to 100 samples is sufficient to achieve a precision of ±10% at
the sites examined.
NRC's current expectations regarding baseline sampling in the ore zone are set out in their
Standard Review Plan for ISR license applications (NRC 2003, p. 5-39), which specifies that one
well per acre should be sampled with four independent samples taken from each well. If a typical
wellfield is 25 acres, then 100 baseline samples would be required. Using the RSEM approach
presented here and assuming that a DQO for the RSEM is fixed at ±10% and the ratio of the
sample mean to the sample standard deviation is 1.0, then the required number of samples would
be 100 or the same as current practice for a typical wellfield. This hypothetical example is
supported by Figure 7-2, which shows, using actual sites, that a baseline sample size of
approximately 80 to 100 samples is sufficient to achieve a precision of ±10% at the sites
examined. If the DQO is reduced to ±5%, the theoretical sample size would be increased to 400.
Any additional sampling could be accommodated by various combinations of additional wells,
extending the sampling period, RSEM or increasing the sampling frequency. However, the
operator must insure that the additional samples are independent in space and time. For example,
if additional baseline sampling wells are required, they should be located in the ore zone as far as
possible from other baseline sampling wells. If more samples are required from each baseline
well, the additional sampling times should be roughly mid-way between current sampling times.
Frequency of sampling should not exceed eight samples per year per well.
1 See Footnote 9 for sources of data.
Draft Technical Report
129
Revised Draft - November 26, 2012
-------
7.1.4 Summary
In practice, the following approach might be taken to determine the number of baseline samples:
• The regulator establishes the DQO for the RSEM (e.g., ± 1 0%).
• The operator collects periodic samples for 1 year based on a sample density of one per
acre per sampling. (From Table 7-4, it is likely that larger wellfields may require
quarterly sampling, while smaller wellfields (e.g., less than 20 acres) may require
monthly sampling.)
The operator calculates the
for each analyte by pooling sampling results.
• Unless the sampling results suggest otherwise, the operator determines whether either
uranium or Ra-226 has an RSEM exceeding the DQO.
• If either uranium or Ra-226 exceeds the DQO, the operator continues sampling until the
DQO is achieved.
This approach is based on the assumptions that uranium and Ra-226 are key analytes in
understanding the ground water chemistry, that they will have few nondetects in the baseline
sampling program, and that, based on the data reported here, they typically have high variability.
Table 7-2. Summary Statistics for Population Coefficient of Variation (cv)
of Baseline Parameters at Nine ISR Sites
Rank
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
Analyte
NO2
COS
Umg/L
NH4
Ra-226 pCi/L
NO3
K
Mg
Ca
F
Cl
HCO3
SiO2
Alk
S04
Cond
TDS
Na
pH
Number of Sites
Major Ions (nig/
2
8
9
7
9
7
9
9
9
9
9
9
9
9
9
9
9
9
9
Minimum cv Average cv Maximum cv
-) and Radiometrics
97% 121% 145%
43% 110% 327%
42% 105% 213%
11% 105% 433%
61% 75% 87%
6% 56% 222%
12% 47% 93%
7% 38% 129%
1% 21% 52%
6% 18% 35%
7% 15% 36%
3% 12% 25%
3% 12% 23%
3% 9% 27%
3% 8% 16%
3% 7% 14%
2% 6% 12%
3% 5% 9%
2% 4% 7%
Draft Technical Report
130
Revised Draft - November 26, 2012
-------
Table 7-2. Summary Statistics for Population Coefficient of Variation (cv)
of Baseline Parameters at Nine ISR Sites
Rank
Analyte
Number of Sites
Minimum cv Average cv Maximum cv
Trace Metals (mg/L)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
Al
Fe
Se
Mn
V
Cr
As
Cd
Hg
Pb
Cu
Ni
Zn
Ba
B
Mo
2
3
7
4
3
1
6
2
1
1
3
2
5
1
2
2
65% 262% 459%
62% 253% 406%
50% 208% 581%
38% 127% 206%
55% 103% 155%
95% 95% 95%
33% 93% 285%
70% 85% 100%
83% 83% 83%
83% 83% 83%
37% 82% 110%
28% 65% 102%
33% 50% 91%
39% 39% 39%
4% 25% 46%
9% 24% 40%
Draft Technical Report
131
Revised Draft - November 26, 2012
-------
Table 7-3. Baseline Statistics and Number of Samples Required at Christensen Mine Unit 6 for the Relative Standard Error of the
Baseline Mean to Be Less Than P% for 35 Analytes with Summary Statistics for U and Ra-226
Analyte
MAJOR IONS
-mg/L
Ca
Mg
Na
K
CO3
HCO3
SO4
a
NH4
NO2
NO3
F
S1O2
IDS
Cond nmho/cm
AlkCaCO3
pH
Number of
Samples*
Total
TV >D.L."
164 164
166 162
168 168
162 162
85 78
85 83
168 168
168 168
86 19
86 0
86 5
85 85
86 86
167 167
167 167
166 166
164 164
Mean*
M
26.238
4.455
239.762
6.61
5.49
74.48
530.643
4.539
0.111
0.076
0.142
4.058
861.09
1261.26
70.753
8.83
Standard
Deviation*
S
4.947
1.417
10.091
2.163
2.968
11.834
40.372
1.091
0.031
0.015
0.049
0.927
50.746
66.99
6.804
0.369
Standard
Error of
the Mean
SEM
0.386
0.110
0.779
0.170
0.322
1.284
3.115
0.084
0.0033
0.0016
0.0053
0.100
3.927
5.184
0.528
0.029
Relative
Standard
Error of the
Mean
RSEM
1.5%
2.5%
0.3%
2.6%
5.9%
1.7%
0.6%
1.9%
3.0%
2.1%
3.7%
2.5%
0.5%
0.4%
0.7%
0.3%
Number of Samples Required for Relative Standard Error of Baseline Mean Less Than P%
P=l% 2% 3% 4% 5% 6% 7% 8% 9% 10% 15% 20% 25% 30% 40% 50% 75% 100%
356 89 40 23 15 10 8 6 5 4 2 222222 2
1012 253 113 64 41 29 21 16 13 11 5 322222 2
18 52222222 2 2 222222 2
1071 268 119 67 43 30 22 17 14 11 5 322222 2
2923 731 325 183 117 82 60 46 37 30 13 854222 2
253 64 29 16 11 8 6 4 4 3 2 222222 2
58 15 7432222 2 2 222222 2
578 145 65 37 24 17 12 10 8 6 3 222222 2
780 195 87 49 32 22 16 13 10 8 4 222222 2
390 98 44 25 16 11 8 7 5 4 2 222222 2
1191 298 133 75 48 34 25 19 15 12 6 322222 2
522 131 58 33 21 15 11 9 7 6 3 222222 2
35 94322222 2 2 222222 2
29 84222222 2 2 222222 2
93 24 11 643 222 2 2 222222 2
18 52222222 2 2 222222 2
TRACE METALS - mg/L
Al
As
Ba
B
Cd
Cr
Cu
Fe
Pb
Mn
Hg
Mo
86 20
167 118
86 0
86 0
86 0
86 1
86 0
86 3
86 0
86 0
86 3
86 7
0.127
0.003
0.05
0.107
0.0003
0.057
0.082
0.001
0
0.066
0
0.005
0.009
0.00008
0.0071
0.00054
7.0%
2.6%
6.7%
0.9%
4169 1043 464 261 167 116 86 66 52 42 19 11 7 5 3 2 2 2
1112 278 124 70 45 31 23 18 14 12 5 322222 2
222222222 2 2 222222 2
3805 952 423 238 153 106 78 60 47 39 17 10 7 5 3 2 2 2
222222222 2 2 222222 2
77 20 9543222 2 2 222222 2
Draft Technical Report
132
Revised Draft - November 26, 2012
-------
Table 7-3. Baseline Statistics and Number of Samples Required at Christensen Mine Unit 6 for the Relative Standard Error of the
Baseline Mean to Be Less Than P% for 35 Analytes with Summary Statistics for U and Ra-226
Analyte
Ni
Se
V
Zn
Number of
Samples*
Total
TV >D.L."
86 4
168 13
86 3
86 21
Mean*
M
0.068
0.0020
0.107
0.015
Standard
Deviation*
S
0.019
0.001
0.059
0.008
Standard
Error of
the Mean
SEM
0.0020
0.00008
0.0064
0.00086
Relative
Standard
Error of the
Mean
RSEM
3.0%
3.9%
5.9%
5.8%
Number of Samples Required for Relative Standard Error of Baseline Mean Less Than P%
P=l% 2% 3% 4% 5% 6% 7% 8% 9% 10% 15% 20% 25% 30% 40% 50% 75%
781 196 87 49 32 22 16 13 10 8 4 222222
2500 625 278 157 100 70 52 40 31 25 12 743222
3041 761 338 191 122 85 63 48 38 31 14 854222
2845 712 317 178 114 80 59 45 36 29 13 854222
RADIOMETRICS
Umg/L
Ra-226 pCi/L
Maximum for
U, Ra-226
P
164 148
168 167
0.0126
105.858
0.0109
90.928
0.00085
7.015
6.8%
6.6%
7484 1871 832 468 300 208 153 117 93 75 34 19 12 9 5 3 2
7379 1845 820 462 296 205 151 116 92 74 33 19 12 9 5 3 2
168 167 - - - 6.8% 7484 1871 832 468 300 208 153 117 93 75 34 19 12 953 2
P=l% 2% 3% 4% 5% 6% 7% 8% 9% 10% 15% 20% 25% 30% 40% 50% 75%
100%
2
2
2
2
2
2
2
100%
a - >D.L. is greater than detection limit
* Source: COGEMA 1996, Table 6, mean is arithmetic mean for all detectable samples
Draft Technical Report
133
Revised Draft - November 26, 2012
-------
Table 7-4. Wellfield Characteristics and Comparison of Actual and Target Baseline Sample Size at
Nine ISR Production Units
Site Acres
Crow Butte MU1 9.3
Highland A 3.0
Highland B 13.8
Irigaray MU1 to MU9 30
Christensen MU2
North 5.7
Christensen MU3 20.4
Christensen MU4 12
Christensen MU5 25
Christensen MU6 42
„ , Number Average Target Maximum
Number
„ of Number Number Target
R ,. Baseline of of Number of
w ,, Wells per Samples Samples Samples for
Acre per Well per Well* RSEM=10%
12 1.3 8.4 8 94
5 1.7 1.0 8 38
20 1.4 - 24 445
46 1.5 4.3 4 162
8 1.4 4.0 9 69
22 1.1 4.0 6 132
12 1 4.0 14 164
25 1 4.0 3 69
42 1 3.9 2 75
U
Actual Achieved
N RSEM
101 9.6%
5 18.6%
-
200 9.0%
32 14.7%
64 14.3%
48 18.5%
99 5.8%
164 6.8%
Target Target
N RSEM
94 10%
18 10%
445 10%
162 10%
69 10%
132 10%
164 10%
34 10%
75 10%
Ra-226
Actual Achieved
N RSEM
101 7.7%
5 27.4%
-
200 7.0%
32 10.8%
64 8.3%
49 11.9%
99 8.3%
168 6.6%
Target Target
N RSEM
60 10%
38 10%
47 10%
76 10%
37 10%
44 10%
70 10%
69 10%
74 10%
Key Sites meeting hypothetical targeted precision goal of 10%.
An alternative is to increase the number of wells.
Draft Technical Report
134
Revised Draft - November 26, 2012
-------
Table 7-5.
Site
Number of Samples Required at Nine Production Units for Relative Standard Error of U and Ra-226 Mean
Baseline Concentrations to Be Less Than ±P%
Wells Samples
per per
Acre Well P=l% 2% 3% 4% 5% 6% 7% 8% 9% 10% 15% 20% 25% 30% 40% 50% 75% 100%
Number of Samples (N) Required for RSEM of U and Ra-226 Baseline Means to Be Less Than ±P%
CrowButteMUl
Actual P and N for (U, Ra)
Highland A
Actual P and N for (U, Ra)
Highland B
Actual P andNfor (U, Ra)
IrigarayMUl-9
Actual P andNfor (U, Ra)
Christensen MU2 North
Actual P andNfor (U, Ra)
Christensen MU3
Actual P andNfor (U, Ra)
Christensen MU4
Actual P andNfor (U, Ra)
Christensen MU5
Actual P andNfor (U, Ra)
Christensen MU6
Actual P andNfor (U, Ra)
1.29
1.67
1.38
1.53
1.40
1.08
1.00
1.00
1.00
8.40
1.00
-
4.35
4.00
4.00
4.00
3.96
3.90
9,359 2,340 1,040 585 375 260 191 147 116 94 42 24 15 11 6 4 2 2
9.6% (101,101)
3,741 936 416 234 150 104 77 59 47 38 17 10 6 5 3 2 2 2
27.4% (5,5)
45,374 11,344 5,042 2,836 1,815 1,261 926 709 561 454 202 114 73 51 29 19 9 5
16,127 4,032 1,792 1,008 646 448 330 252 200 162 72 41 26 18 11 7 3 2
9% (200,200)
6,877 1,720 765 430 276 192 141 108 85 69 31 18 12 8 5 3 2 2
14. 7% (3 '2,3 '2)
13,142 3,286 1,461 822 526 366 269 206 163 132 59 33 22 15 9 6 3 2
14. 3% (64,64)
16,352 4,088 1,817 1,022 655 455 334 256 202 164 73 41 27 19 11 7 3 2
18.5% (48,49)
6,838 1,710 760 428 274 190 140 107 85 69 31 18 11 8 5 3 2 2
8.3% (99,99)
7,484 1,871 832 468 300 208 153 117 93 75 34 19 12 9 5 3 2 2
6.8% (164,168)
Summary Statistics
Across All Sites
Average
Minimum
Maximum
Wells
per
Acre
1.24
1.00
1.67
Samples
per
Well
4.18
1.00
8.40
Number of Samples (N) Required for Coefficients of Variation of U and Ra-226 Baseline Means to Be Less Than ±P%
P=l% 2% 3% 4% 5% 6% 7% 8% 9% 10% 15% 20% 25% 30% 40% 50% 75% 100%
13,922 3,481 1,547 870 557 387 285 218 172 140 62 35 23 16 9 6 3 2
3,741 936 416 234 150 104 77 59 47 38 17 10 6 5 3 2 2 2
45,374 11,344 5,042 2,836 1,815 1,261 926 709 561 454 202 114 73 51 29 19 9 5
Draft Technical Report
135
Revised Draft - November 26, 2012
-------
30% -i
^ 25%
g 20%
15%
"0 10%
x "5 5%
0%
50 100 150 200
Baseline Sample Size
250
Figure 7-2. Scatter Plot of the Maximum RSEM for U and Ra-226 Mean Concentrations
versus Baseline Sample Size at Nine Production Units
7.2 Determining the Number of Monitoring Wells Required to Detect Noncompliance
This section presents a statistical approach to determining the number of post-operational
monitoring wells. The number of wells to monitor in the post-operational period will depend on
the size of the field and the connectivity (i.e., the degree to which the wells are hydraulically
connected) of the wells. One approach to this question involves the theory of hypergeometric
sampling. The monitoring wells discussed in this section are wells within the ore zone and not
the perimeter monitoring wells used to detect excursions discussed in Section 3.3.1.
If the wells in the production unit have high connectivity, then fewer monitoring wells would be
required to achieve adequate coverage. A modified approach for wellfields with some measure of
connectivity is discussed in Section 7.2.2. In practice, it may prove difficult to quantify the
qualitative measures of connectivity discussed in that section. If that proves to be the case, the
hypergeometric sampling approach presented in the following section is most appropriate.
7.2.1 Determining the Number of Monitoring Wells based on Hypergeometric Sampling
Experience has shown that it is unlikely that every monitored well in a production unit will be in
compliance. Hypothesis tests are used to determine if the monitored wells have returned to
baseline conditions and have achieved steady state. Tests for comparing the data collected at the
site after restoration with baseline conditions are discussed in Section 7.3. Tests for trends are
discussed in Section 7.7. We assume that the test employed is adequate for the task of
determining which of the monitored wells are in compliance and which are not. Every test has a
"gray region" where the test may have relatively high decision error rates, but these tests perform
with almost 100% accuracy when the monitored wells are not near the gray region. Although the
tests are performed only on the monitored wells, we are interested in extrapolating the test results
Draft Technical Report
136
Revised Draft - November 26, 2012
-------
obtained for the monitored wells to other wells in the production unit, assuming they would have
a similar pattern of test outcomes if all wells were monitored.
If the test outcomes indicate that all monitored wells demonstrate compliance, or if only a few do
not, then these results also provide information about the wells which were not monitored.
Suppose the question is phrased in this manner:
If all wells were monitored, what percentage of the wells would be expected to not
demonstrate compliance?
If all wells were monitored, this percentage would be known. When fewer than all wells are
monitored, the percentage can be known only with uncertainty. The statistical approach involves
confidence levels for the unknown parameter Q defined as the percentage of wells that would not
demonstrate compliance if all wells were monitored16 The compliance determinations for the
monitored wells are based on statistical tests that will have Type 1 and Type 2 errors. Hence, a
test indicating that a monitored well is in compliance does not mean that the well truly is in
compliance, and a test indicating that a monitored well is not in compliance does not mean that
the well truly is not in compliance. This limitation also is applicable for Q, which is a measure of
the proportion of wells with test results that are not in compliance.
In the following discussion, the set of all wells in the production unit form the "population" from
which a "sample" of monitored wells is selected. In many sampling problems, it is assumed that
the population to be sampled is much larger than the intended size of the sample. Essentially, the
population size is assumed to be of infinite size. When a population of finite size is sampled, the
required size of the sample is less than the corresponding sample size for an infinite population.
In a population with a relatively small size (in the hundreds), the savings can be significant. With
a population of finite size, it is theoretically possible to determine exactly any population
parameter of interest by monitoring all of the wells. When only a sample of wells is monitored,
then population parameter estimates will have uncertainty and a statistical approach is used for
estimating these parameters.
There are several statistical approaches for estimating the population parameter Q. Two
parametric statistical approaches are the traditional frequentist statistical approach, such as
maximum likelihood estimation and the Bayesian approach.
In maximum likelihood estimation, the goal is to find the "best" point estimate for Q, defined as
the value of Q which maximizes the likelihood function /(Q, X) for the observed set of sample
values X=(xi, ..., xn). This estimate is a "statistic," meaning that it depends on the sample values
X. The most commonly used maximum likelihood estimate is the sample mean. Since this
estimate is a function of the sample values, the estimate would vary if the sampling process was
16 The percentage of wells that are not in compliance is an observable parameter that serves as a surrogate
for the question of greater interest: What percentage of the underlying aquifer is not in compliance? The interlaced,
5-spot and 7-spot patterns of injection/extraction wells serve as an efficient sampling grid for monitoring the
underlying aquifer(s). If geostatistical software is available, indicator kriging may be used to estimate the percent of
compliance throughout the unit.
Draft Technical Report 137 Revised Draft - November 26, 2012
-------
repeated. Confidence intervals for the "true" value of Q are formed by calculating the frequency
of values of Q that would occur when the sampling process is repeated.
In the Bayesian approach, it is assumed that the value of the population parameter Q is always
uncertain, both before and after the sampling process. It is also assumed that this uncertainty can
be expressed in the form of a probability distribution for the unknown parameter Q. This
probability distribution expresses the uncertainty in the estimate of Q. If Q is relatively well
known, then the probability distribution is a narrow distribution, but if the value of Q is very
uncertain, then the probability distribution is wide.
Before the sampling is done, the prior state of knowledge about the parameter Q is expressed in a
prior distribution written as/>(Q). After the sampling is completed, the data are used to update
the prior state of knowledge about the parameter Q (using Bayes' theorem) to a new, narrower
distribution called the posterior distribution p(Q\X). The posterior distribution is proportional to
the product of the prior distribution times the likelihood function: XQI20 ~XQ) AQ, 20
In this application, a non-informative prior distribution is used for Q. The only prior knowledge
assumed is that Q must be between 0 and 1, and no value of Q in between these limits is more
likely that any other value. These assumptions imply that the prior distribution for Q is a uniform
distribution between 0 and 1, i.e., p(Q)=l for all Q. Note that/?(Q|X) ~ AP, 20 with this choice
of a prior distribution. The value of Q, which maximizes the likelihood function, is the same
value that maximizes the Bayesian posterior distribution for Q. Hence, the maximum likelihood
estimate is equal to the Bayesian highest posterior density estimate using a non-informative prior
distribution.
In addition to the point estimate generated using the Bayesian approach, both 1-sided and 2-sided
intervals of highest posterior density (HPD) may be constructed containing 90%, 95% or any
desired level of posterior probability. The shape of the posterior distribution is, in general, not
symmetric around the highest posterior density point estimate. This means that the HPD intervals
for Q constructed using the Bayesian approach are in general asymmetric. The degree of
asymmetry increases at values of Q near 0 or 1. The HPD intervals are approximately symmetric
when Q is near l/2.
In this section, it is assumed that a sufficient number of samples have been collected from each
monitored well to determine if the well demonstrates compliance. (Sections 7.6 and 7.8 discuss
the procedures for determining compliance of an individual well.) After the sampling is
completed, some of the monitored wells will have demonstrated compliance and others will not.
These results are used to determine the maximum likelihood (and Bayesian) point estimate of Q
equal to the percentage of monitored wells not demonstrating compliance:
Q = (lOO)(# of monitoredvellsnotshowing:omplianc^(N)
This point estimate of Q will have statistical uncertainty. The required number of monitoring
wells is determined by placing an upper bound on the posterior distribution for Q. If there areM
wells in a production unit, then a sufficient number TV are designated as monitoring wells, such
that the Bayesian 1-sided highest posterior density interval for Q is within the specified bounds.
Draft Technical Report 13 8 Revised Draft - November 26, 2012
-------
If all monitored wells are found to demonstrate compliance, then the point estimate for Q is 0%.
However, this estimate also has uncertainty, as there still may be unmonitored wells that would
be expected not to demonstrate compliance. If a well production unit has only 5 wells, then
monitoring of all wells may be required to achieve the desired level of confidence in the estimate
of Q. If there are 200 wells in a production unit, it will be necessary to monitor more than 5
wells, but not to monitor all the wells. However, if the number of noncomplying wells in the unit
is small relative to M, it is very difficult to detect that there are wells out of compliance without
very extensive monitoring.
A Bayesian analysis of the hypergeometric distribution is used to determine the number of
monitoring wells required to demonstrate, with a relatively high degree of assurance, that no
more than q percent of the wells in the field would not demonstrate compliance if all wells were
monitored. Let X denote the number of monitored wells that do not demonstrate compliance in
the set of TV monitoring wells. Let Y denote the unknown number of wells in the field that would
not show compliance if all wells were tested. Then the random variable Xhas a hypergeometric
distribution with parameters TV, M and Y. A realization of the random variable Xis a nonnegative
integer denoted by the lower case symbol x. Table 7-6 shows the parameters of the
hypergeometric distribution and the range of values for each parameter.
Table 7-6. Parameter Definitions for the Hypergeometric Distribution
Parameter
Number of Wells
Number of Wells not
Showing Compliance
All Wells in Production Unit
(Population)
M
M>0
Y
x b, and the symbol a\ denotes the
\.b)
factorial of the integer a = (1)(2)(3).. .(a). By convention, 0! = 1. Note that the term (M-y) is the
number of wells in the production unit that would demonstrate compliance if all wells were
tested, and (N-x) is the number of monitored wells that demonstrate compliance.
For ease of discussion, we denote a test outcome that demonstrates compliance as a "Pass;"
otherwise, the test outcome is a "Fail." The posterior distribution for the (unknown) number of
Draft Technical Report 13 9 Revised Draft - November 26, 2012
-------
wells 7 in the population with outcome "Fail," given that x wells had the outcome "Fail" in the
sample of TV wells, is:
y\
P[Y=y\x,M,N] = y'
(y-X)\[(M-y)-(N-X)]\
with k = l^P[Y = y\x,M,N].
Here 7 is a random variable with integer values^ = x, ... , M - (N - x), the posterior distribution
P[Y = y x,M,N] gives the probability of each possible value of 7, and k is a normalizing
constant such that these probabilities add to one. The parameter Q is the unknown percentage of
wells in the population that would have an outcome of "Fail" if all wells had been monitored. For
each field sizeMthere areM+1 discrete values possible for Q, one for each possible value of 7:
Q = 100j7M,j = (),••• ,M . The posterior distribution for Q is derived from the posterior
distribution for 7 by linear transformation.
The posterior distribution for Q also depends on x, the number of Fails observed in the monitored
wells. It is possible to achieve the desired upper bound on Q using several different monitoring
strategies. The strategies are characterized by the upper bound on the number of monitored wells
permitted to Fail while still meeting the desired bound on Q. The bound can be achieved by
monitoring a smaller number of wells with the requirement that all monitored wells must pass
(x=0) to achieve the desired bound on Q. Alternatively, by monitoring a larger number of wells,
the same bound on Q can be achieved while allowing for one of the monitored wells to Fail
(x=l). The posterior distribution for Q was used to calculate one-sided highest posterior density
intervals for Q of the form 0 < Q < q. Table 7-7 and Table 7-8 are derived from the posterior
distribution for Q. Table 7-7 is used when all of the monitored wells must Pass, and Table 7-8 is
used when one monitored well is permitted to Fail. The number of monitoring wells shown in
Table 7-8 is higher than in Table 7-7. Additional tables for obtaining posterior probabilities of
90% and 99% with x=0 or x=l, and tables for achieving posterior probabilities of 90%, 95% and
99% in the case x=2, are found in Tables E-15 through E-21 of Attachment E, respectively.
When all monitored wells must "Pass" (x=0), Table 7-7 shows the minimum number of
monitoring wells N, such that the probability that Q is less than q is at least 95%, i.e.,
Prob{0 < Q < q | x = 0, M, N} > 0.95 .
If it is determined that all of the TV monitored wells "Pass" (i.e., x=0), then there is less than a 5%
chance that more than q percent of the wells in the production unit would "Fail" if all M wells
were monitored.
For example, consider a 5-spot wellfield with approximately 200 wells. WithM=200 wells,
Table 7-7 indicates that 25 monitoring wells are required for 95% posterior probability that Q is
less than 10% if all monitored wells Pass. If 46 wells are monitored and all Pass, there is a 95%
probability that Q is less than 5%. If only 17 wells are monitored and all Pass, there is a 95%
probability that Q is less than 15%.
Draft Technical Report 140 Revised Draft - November 26, 2012
-------
Table 7-8 shows similar results for the case when one of the monitored wells is permitted to Fail
Prob{0 < Q < q | x = 1, M, N} > 0.95 .
If one of the monitored wells is permitted to Fail in the example withM=200 wells, the N=39
monitoring wells would be required to have 95% probability that Q is less than 10%.
As noted above, Q is an unknown parameter defined as the percentage of wells that would not
demonstrate compliance if all wells were tested. In order to choose between Table 7-7 and Table
7-8, an estimate of x, the number of monitored wells expected to Fail, is needed. This estimate
can be obtained from results of post-restoration sampling, which will likely be done on the
baseline wells. The number of baseline wells is currently specified as one well per acre (NRC
2003). If the wellfield has an average area of 25 acres,1 baseline samples and the first round of
post-remediation samples would be obtained from 25 wells. Post-restoration sampling of these
wells could then be used to estimate x. If all wells demonstrate compliance, then Table 7-7 can
be used to determine the number of monitoring wells required to demonstrate that the restored
wellfield is in compliance. If x=l, then Table 7-8 should be used. Additional tables for x=2 are
included in Attachment E. If x>2, additional restoration would likely be required before
proceeding to demonstrate compliance.
To determine the number of wells that must be sampled to establish wellfield compliance, DQOs
must be set for 0, which is the upper bound for Q, and for the probability that Q will not exceed
q. If the DQOs are 0=5% and Probjg < 5%} > 0.95, then using Table 7-7 one can estimate that
52 wells must be monitored in a 25-acre wellfield, with 375 production plus injection wells to
satisfy the DQOs. This number is approximately twice the number of baseline wells, so an
expanded number of monitoring wells would be required in the post-restoration period.
However, if the DQOs are 0=10% and Probjg < 10%} > 0.95, then 26 wells would require
monitoring. This number is approximately equal to the number of baseline wells. Each of the 26
wells in this example would be tested for compliance. If the compliance testing results indicated
all of the 26 monitoring wells demonstrate compliance, then there is a 95% probability that no
more than 10% of the wells (or, by extension, roughly 10% of the wellfield ground water) would
not demonstrate compliance if all wells were tested. If one of the monitored wells is permitted to
not demonstrate compliance, then using the Table 7-8 DQOs of 0=10% and Prob{£> < 10%} >
0.95 can be achieved by monitoring 42 wells.
The number of samples required per well during the post-restoration sampling period is
discussed in Section 7.6.
17 Based on a survey of a number of ISL sites, an average wellfield was 25 acres and contained about 15
wells per acre, for a total of 375 wells.
Draft Technical Report 141 Revised Draft - November 26, 2012
-------
able 7-7.
Minimum Value of N with Prob {Q < q
(Assumes No Failures in TV Monitored W<
x=0,
3lls)
M,N}>
0.95
q (%)
3
.
-
-
-
-
-
-
-
-
-
-
-
19
24
29
27
31
35
39
43
46
50
44
47
50
4
.
-
-
-
-
-
-
-
-
-
-
-
19
19
23
27
31
35
31
34
37
41
44
39
41
5
.
-
-
-
-
-
-
-
-
-
-
-
15
19
23
27
25
28
31
34
31
33
36
39
35
6
.
-
-
-
-
-
-
-
-
13
14
15
15
19
23
22
25
28
26
28
31
33
31
33
35
7
.
-
-
-
-
-
-
12
12
13
14
15
15
19
18
22
25
23
26
28
26
28
31
28
30
8
.
-
-
-
-
10
11
12
12
13
14
15
15
15
18
22
20
23
22
24
26
24
26
25
27
9
.
-
-
-
9
10
11
12
12
13
14
15
15
15
18
18
20
19
22
24
22
24
23
25
24
10
.
-
8
8
9
10
11
12
12
13
14
15
12
15
15
18
17
19
19
20
20
21
21
22
21
11
.
-
8
8
9
10
11
12
12
13
14
11
12
15
15
18
17
19
19
18
20
19
21
20
21
12
.
7
8
8
9
10
11
12
12
10
11
11
12
12
15
15
17
17
16
18
17
19
18
18
19
13
6
7
8
8
9
10
11
12
10
10
11
11
12
12
15
15
15
17
16
16
17
17
17
18
18
14 15 16
6 6
7 7
8 8
8 8
9 9
6
7
8
8
9
10 10 8
11 8
9 9
8
9
10 10 10
10 10 10
11 11 11
11 11 9
12 10 10
12 12 10
13 13 13
15 13 13
15 13 13
14 14 13
16 14 13
16 14 14
16 14 14
15 15 14
17 15 14
16 15 14
16 15 15
17
4
T-
6
7
8
8
7
8
8
9
10
10
9
9
10
10
11
13
13
13
13
13
13
13
14
14
14
18
4
T-
6
7
8
8
7
8
8
9
10
8
9
9
10
10
11
11
11
11
11
13
13
13
13
13
13
19
4
*T
6
7
8
6
7
8
8
9
8
8
9
9
10
10
11
11
11
11
11
12
12
12
12
12
12
20
4
4
T-
6
7
6
6
7
8
8
7
8
8
9
9
8
9
9
10
10
10
10
11
11
11
11
11
11
25
4
4
T-
4
5
6
6
5
6
7
7
6
7
7
8
7
7
8
8
8
8
9
9
9
9
9
9
9
30
4
4
T-
A
T-
4
5
4
5
5
6
5
6
6
5
6
6
6
6
6
7
7
7
7
7
7
7
7
7
7
40
2
J
A
T-
3
4
3
4
4
4
4
4
4
5
4
5
4
4
5
5
5
5
5
5
5
5
5
5
5
50
2
9
Z,
2
3
3
3
3
3
3
3
3
3
3
3
3
3
3
4
3
4
4
4
4
4
4
4
4
-------
Table
M
90
95
100
110
120
130
140
150
160
170
180
190
200
225
250
300
350
400
450
500
600
700
800
900
1000
7-7.
Minimum Value of N with Prob {Q < q
(Assumes No Failures in TV Monitored W<
x=0,
3lls)
M,
N}>
0.95
Q (%)
0
86
91
95
105
114
124
133
143
152
162
171
181
190
214
238
285
333
380
428
475
570
665
760
855
950
1
86
91
78
85
93
101
109
116
124
132
140
147
126
142
157
157
184
179
202
195
208
217
225
232
237
2
70
74
63
69
75
82
88
78
84
89
94
99
89
100
97
103
108
112
115
118
122
125
128
130
131
3
56
59
52
57
62
68
62
67
71
66
70
74
68
77
77
76
82
81
85
84
86
87
89
90
90
4
47
49
44
49
53
50
54
51
54
58
55
58
55
57
58
60
62
63
64
65
66
67
68
68
69
5 6 7 8 9 10 11 12 13
40 34 30 27 24 22 22 20 18
42 36 32 28 26 23 21 19 18
38 34 34 27 24 22 21 19 18
42 37 33 30 27 23 21 20 18
41 36 33 30 27 23 22 20 19
44 39 32 30 27 23 22 21 19
42 38 35 29 27 24 22 21 19
46 37 34 29 27 24 23 20 19
44 40 34 31 27 24 23 21 20
47 39 36 31 27 24 23 21 19
45 41 35 31 27 25 23 21 19
48 40 35 31 27 25 24 21 20
46 40 37 31 27 25 23 21 19
48 42 37 31 28 26 24 21 20
50 41 37 32 29 25 24 21 20
50 42 38 32 29 26 23 21 20
52 43 38 33 29 26 24 22 20
51 43 39 33 29 26 24 22 20
53 44 38 33 30 26 24 22 20
53 44 39 33 30 27 24 22 20
53 45 39 34 30 27 24 22 20
54 45 40 34 30 27 24 22 20
55 46 40 34 30 27 25 22 21
55 46 40 34 30 27 25 22 21
55 46 40 34 30 27 25 22 21
14 15 16
17 16 15
17 16 15
18 15 14
17 16 15
18 16 15
17 16 16
18 16 15
18 17 15
18 16 16
18 17 15
18 16 16
18 17 16
18 17 16
18 17 16
19 17 16
19 17 16
19 17 16
19 17 16
19 17 16
19 17 16
19 17 16
19 18 16
19 18 16
19 18 16
19 18 16
17
14
14
14
14
14
14
15
15
14
15
15
15
15
15
15
15
15
15
15
15
15
15
15
15
15
18
13
13
13
14
14
13
13
13
14
14
14
14
14
14
14
14
14
14
14
14
14
14
14
14
14
19
12
12
12
13
13
13
13
13
13
13
13
13
13
13
13
13
13
13
13
13
13
13
14
14
14
20
11
11
12
12
12
12
12
12
12
12
12
12
12
12
12
12
12
13
13
13
13
13
13
13
13
25
9
9
9
9
9
9
9
9
9
10
9
10
9
10
10
10
10
10
10
10
10
10
10
10
10
30
7
7
7
7
7
7
7
8
8
8
8
8
8
8
8
8
8
8
8
8
8
8
8
8
8
40
5
5
5
5
5
5
5
5
5
5
5
5
5
5
5
5
5
5
5
5
5
5
5
5
5
50
4
4
4
4
4
4
4
4
4
4
4
4
4
4
4
4
4
4
4
4
4
4
4
4
4
Draft Technical Report
143
Revised Draft - November 26, 2012
-------
Table
7-8.
Minimum Value
(Assumes
M
7
17
16
17
18
19
20
25
30
35
40
45
50
55
60
65
70
75
80
85
of N with Prob
{Q
0.95
1 Failure in TV Monitored Wells)
Q (%)
0123
_
.
.
.
.
.
.
34
38
43
48 48
53 53
57 57
62 62
67 56
72 60
76 64
81 68
4 5
_
-
-
-
- 19
24 24
29 29
34 34
38 32
43 36
40 40
44 44
48 40
52 44
56 47
51 51
54 46
57 49
6
_
-
-
-
19
24
29
28
32
36
34
37
40
44
40
43
46
43
7 8
_
-
-
-
19 19
24 20
24 24
28 28
32 27
30 30
34 29
37 32
34 34
37 32
40 35
37 33
40 35
43 38
9
_
-
-
-
19
20
24
23
27
26
29
32
30
32
31
33
31
33
10 11
_
-
-
- 15
16 16
20 20
20 20
23 23
23 23
26 26
25 25
27 24
26 26
28 25
27 27
29 26
28 28
30 27
12 13
- 13
14 14
14 14
15 15
16 16
17 17
20 20
20 20
23 20
22 22
22 22
24 21
23 23
25 23
25 22
24 24
26 24
25 23
14 15
19 19
13 13
14 14
14 14
15 15
16 13
17 17
17 17
20 17
20 17
19 19
22 19
21 19
21 19
21 21
22 20
22 20
22 20
23 21
16
ID
19
13
14
14
13
13
14
17
17
17
17
17
19
19
19
19
19
20
20
17 18
10 10
ID ID
Ui i
19 19
13 13
14 11
12 12
13 13
13 13
14 14
14 14
17 15
17 15
17 15
17 16
17 17
17 17
17 17
19 17
19 17
19 17
19 17
19 20
9 9
10 10
ID ID
Ui i
19 10
10 10
11 11
12 12
13 13
13 11
14 12
14 13
15 13
15 14
15 14
16 14
16 14
16 15
16 15
16 15
16 15
16 15
16 15
25 30
6
6 6
7 7
« f.
9 7
8 8
c c
q c
10 8
9 9
9 8
10 8
10 9
9 8
10 9
11 9
12 10
11 9
12 10
12 10
12 10
12 10
12 10
12 10
13 10
12 10
12 10
40
4
f.
f.
6
7
6
7
6
6
7
7
7
7
7
7
7
7
7
7
7
7
50
4
4
4
4
•s
4
5
4
-------
Table
7-8.
Minimum Value
(Assumes
M
90
95
100
110
120
130
140
150
160
170
180
190
200
225
250
300
350
400
450
500
600
700
800
900
1000
of N with Prob
{Q
0.95
1 Failure in TV Monitored Wells)
Q (%)
0 1
-
-
- 95
- 105
- 114
- 124
- 133
- 143
- 152
- 162
- 171
- 181
- 161
- 181
- 201
- 204
- 237
- 233
- 262
- 254
- 271
- 283
- 294
- 302
- 309
2
86
91
80
88
96
105
113
102
108
115
122
129
116
131
127
135
141
146
150
154
159
163
167
169
171
3
72
76
68
74
81
88
81
87
93
86
91
96
90
101
101
100
107
106
111
110
112
114
116
117
118
4 5
61 52
64 55
58 50
64 55
69 53
66 58
71 56
67 60
71 58
76 62
72 59
76 63
73 61
75 63
76 66
79 65
81 68
83 68
84 70
85 69
86 70
88 71
89 72
89 72
90 72
6
45
48
44
49
48
52
50
49
53
51
55
53
52
55
54
55
56
57
58
58
59
59
60
60
60
7 8
40 36
42 38
44 36
44 39
43 39
43 39
46 39
45 39
45 41
48 41
47 41
46 41
49 41
49 41
48 42
50 42
50 43
51 43
51 44
52 44
52 44
52 45
52 45
52 45
53 45
9
32
34
32
36
36
36
36
36
36
36
36
36
36
37
38
38
39
39
39
39
40
40
40
40
40
10 11
29 29
31 28
30 27
30 28
31 29
31 29
32 30
32 30
32 30
32 31
33 31
33 31
33 30
34 31
34 31
34 31
35 32
35 32
35 32
35 32
36 32
36 32
36 33
36 33
36 33
12 13
27 25
26 24
25 24
26 24
27 25
27 26
28 25
27 26
27 26
28 25
28 26
29 26
28 26
28 26
28 26
29 26
29 27
29 27
29 27
29 27
29 27
30 27
30 27
30 27
30 27
14 15
23 21
22 21
24 21
23 22
24 21
23 22
24 22
24 22
24 22
24 22
24 22
24 23
25 22
24 23
25 23
25 23
25 23
25 23
25 23
25 23
25 23
25 23
26 23
26 24
26 24
16
20
20
20
20
20
21
21
20
21
21
21
21
21
21
21
21
21
22
22
22
22
22
22
22
22
17 18
19 17
18 17
18 17
19 18
19 18
19 18
20 18
20 18
19 19
20 19
20 19
20 19
20 18
20 19
20 19
20 19
20 19
20 19
20 19
20 19
20 19
20 19
21 19
21 19
21 19
19
16
16
17
17
17
17
17
17
17
17
17
17
17
18
18
18
18
18
18
18
18
18
18
18
18
20 25
16 13
16 13
16 12
16 13
16 13
16 13
16 13
16 13
16 13
16 13
16 13
16 13
17 13
17 13
17 13
17 13
17 13
17 13
17 13
17 13
17 13
17 13
17 13
17 13
17 13
30
10
10
10
10
10
10
10
10
10
10
11
11
11
11
11
11
11
11
11
11
11
11
11
11
11
40 50
7 5
7 6
7 5
7 5
7 5
7 6
7 6
7 6
7 6
7 6
7 6
7 6
7 6
7 6
8 6
8 6
8 6
8 6
8 6
8 6
8 6
8 6
8 6
8 6
8 6
Draft Technical Report
145
Revised Draft - November 26, 2012
-------
7.2.2 An Alternative Graduated Approach to Hypergeometric Sampling
This section describes a series of graduated sampling design options for production units with
varying degrees of hydraulic connectivity. This approach assumes that some measure of
connectivity between wells is available as discussed further in Section 7.2.3. In each design
option, the production unit is divided into M clusters of wells. The clusters, all of similar size and
sufficient in number to cover the entire production unit, are called "well zones" in this document.
The intent is to monitor a sufficient number of well zones (TV) to ensure a high probability that
most zones in the production unit would be in compliance if all monitored zones are found to be
in compliance. In the graduated approach, production units with higher connectivity have larger
well zones and require fewer monitoring wells to achieve the desired probability levels.
Each design option is based on a two-stage sampling strategy. In the first stage, TV well zones are
selected for monitoring. In the second stage, one monitoring well is randomly selected in each of
the well zones chosen in the first stage. This two-stage procedure generates a set of TV monitoring
wells, which will be used to determine the compliance rate in the entire production unit. In this
section, it is assumed that all N monitoring wells are to be in compliance, with no failures.
Table 7-9 shows the data collected for 15 production units at four ISR sites. On average, there
are 181 wells in a typical production unit, comprising approximately 100 injection wells (56%)
and 80 production (or extraction) wells (44%). In the following discussion, the production unit is
assumed to be contiguous, use a five-spot pattern, and the ore zone is assumed to have a square
shape. Similar design alternatives for production units using a seven-spot pattern are also
considered.
Figure 7-3 through Figure 7-7 show five sampling design options for a 5-spot production unit for
units with minimal, low, moderate, good and high connectivity. Each figure shows the 181 wells
in the production unit and a checker board pattern of well zones of varying size. Although these
examples assume rectangular ore zones with rectangular well zones, the design options may be
adapted to fields with irregular geometries like the one shown in Figure 7-1 using irregularly
shaped clusters all of approximately the same number of wells. In each figure, the well zones
contain an average of A: wells. A design with dwells in each well zone is described as a "k-well
design'' Production units with higher connectivity have larger well zones and a larger value of
k.
Figure 7-3 presents a two-well design for use in production units with minimal connectivity.
Using this design, there are 81 well zones and each zone contains 1 production well. In the first
stage, a sample of TV of the 81 well zones is selected. In the second stage, one of the k=2 wells in
each zone is randomly designated as the compliance monitoring well for that zone. The required
number of well zones to be monitored (TV) provides a high posterior probability that a large
percentage (P) of all 81 zones in the production unit would be found in compliance if all
monitored zones are found to be in compliance.
Draft Technical Report 146 Revised Draft - November 26, 2012
-------
Table 7-10 shows the number of monitored zones (N) required to achieve a posterior probability
of 90% that at least P percent18 of the zones are in compliance for values of P ranging from 75%
to 99%. For example, to achieve a posterior probability of 90% that at least P=95% of the zones
are found in compliance using the 2-well design, 29 post-operational monitoring wells are
required. Table 7-11 is used for a posterior probability of 95%. To achieve a posterior probability
of 95% that at least ^=95% of the well zones are found in compliance, 35 monitoring wells are
required. Tables 7-10 and 7-11 are derived from Table 7-7, and Table E-15 using P=1-Q. Note
that Tables 7-7 and E-15 are based on the assumption that all monitored wells are found in
compliance (x=0).
Figure 7-4 shows a 5-well sample design used for production units with low connectivity. The 5-
well design contains 36 well zones and the zones have an average of 5 wells. A sample of TV of
the 36 well zones is selected in the first stage, and then 1 of the wells in each selected zone is
randomly designated as the compliance monitoring well. In this design, 24 monitoring wells are
required to achieve 90% confidence that at least ^=95% of the well zones are in compliance. To
achieve 95% confidence that at least 95% of the well zones are in compliance requires 27
monitoring wells.
Figure 7-5 shows a 7-well design used for production units with moderate connectivity. The
7-well design contains 25 well zones with an average of approximately 7 wells in each zone. As
shown in Table 7-10, 17 monitoring wells are required to achieve 90% confidence that at least
^=95% of the well zones are in compliance. To achieve 95% confidence that at least 95% of the
well zones are in compliance requires 20 monitoring wells.
Figure 7-6 and Figure 7-7 show the 11-well and 20-well design options used for production units
with good or high connectivity, respectively. The 11-well design contains 16 well zones with an
average of approximately 11 wells in each zone. The 20-well design contains 9 well zones with
an average of approximately 20 wells in each zone. Based on Table 7-10, 15 monitoring wells
are required in the 11-well design to achieve 90% confidence that at least 95% of the well zones
are in compliance. In the 20-well design, all 9 production units must be monitored to achieve the
90% confidence level. To achieve 95% confidence that at least 95% of the well zones are in
compliance, all well zones must be monitored when using the 11-well or 20-well designs.
Figure 7-8 through Figure 7-11 show 4 sampling design options for 7-spot production units for
units with minimal, low, good and high connectivity. Figure 7-8 shows a 2-well design with 72
well zones and an average of 2.25 wells per zone. Figure 7-9 describes a 4-well design with 36
well zones and an average of 4.5 wells per zone. Figure 7-10 shows a 10-well design with 16
well zones and an average of 10 wells per zone. Figure 7-11 shows an 18-well design with 9 well
zones and an average of 18 wells per zone.
Table 7-11 shows the value of TV required to achieve 90% confidence that at least/1 percent of
the well zones are in compliance when using a 5-spot pattern. Table 7-13 is used for a 90%
confidence level with 7-spot patterns.
18 In the previous section, the parameter Q was used to denote the percentage of wells that would not be in
compliance if all wells were monitored. In this section, it is more convenient to use the parameter P = 100-Q. P is
the percentage of wells that would be found in compliance.
Draft Technical Report 147 Revised Draft - November 26, 2012
-------
Table 7-10 and Table 7-11 are useful only for production units with 181 wells, and Table 7-12
and Table 7-13 for 162 wells. For larger or smaller production units with different shapes, the
number of well zones determined using a &-well design will vary. Table 7-14 and Table 7-15
contain the information required to construct Table 7-10 through Table 7-12 with design options
for production units with a different number of well zones and different geometries. Geometries
with disjointed clusters of wells require a separate well zone for each cluster.
To compare the number of monitoring wells required in this section with connectivity of varying
degrees and in the previous section with no assumed connectivity, consider again the 5-spot unit
with 181 wells. Table 7-11 shows that 35, 27, 19, 16, or 9 monitoring wells are required for a
posterior probability of 95% that at least 95% of the wells are expected to be in compliance using
a 2-, 5-, 7-, 11- or 20-well design, respectively. In comparison, 45 wells are required in Table 7-7
for a posterior probability of 95% that the percent of wells expected to not be in compliance is
less than 5% in a unit with no connectivity. With no connectivity, 45 wells are designated as
monitoring wells. If minimal connectivity is assumed and a 2-well design adopted, the required
number of monitoring wells is reduced to 35. If a 5-well design is adopted, the required number
of monitoring wells is reduced further to 27. Additional reductions in the required number of
monitoring wells are obtained if a higher degree of connectivity is demonstrated to support a
7-well or 11-well design.
Draft Technical Report 148 Revised Draft - November 26, 2012
-------
Production Unit with 5-spot pattern, 100 injection wells (•) and 81 production wells (0).
Each production well defines a well zone. Unit has 81 well zones (shading).
Figure 7-3. 2-Well Design for Production Unit with Minimal Connectivity
.
Production Unit with 5-spot pattern, 100 injection wells (•) and 81 production wells (0).
Unit has 36 well zones (shading) and an average zone contains 5 wells.
Figure 7-4. 5-Well Design for Production Unit with Low Connectivity
Draft Technical Report
149
Revised Draft - November 26, 2012
-------
.
0
Production Unit with 5-spot pattern, 100 injection wells (•) and 81 production wells (0).
Unit has 25 well zones (shading) and an average zone contains 7 wells.
Figure 7-5. 7-Well Design for Production Unit with Moderate Connectivity
Production Unit with 5-spot pattern, 100 injection wells (•) and 81 production wells (0).
Unit has 16 well zones (shading) and an average zone contains 11 wells.
Figure 7-6. 11-Well Design for Production Unit with Good Connectivity
:
Draft Technical Report
150
Revised Draft - November 26, 2012
-------
•
»
Production Unit with 5-spot pattern, 100 injection wells (•) and 81 production wells ({°}).
Unit has 9 well zones (shading) and an average zone contains 20 wells.
Figure 7-7. 20-Well Design for Production Unit with High Connectivity
.
Draft Technical Report
151
Revised Draft - November 26, 2012
-------
Production Unit with 7-spot pattern, 126 injection wells (•) and 36 production wells ({°}).
Unit has 72 well zones (shading) and an average zone contains 2 wells.
Figure 7-8. 2-Well Design for Production Unit with Minimal Connectivity
Draft Technical Report
152
Revised Draft - November 26, 2012
-------
Production Unit with 7-spot pattern, 126 injection wells (•) and 36 production wells ({°}).
Each production well defines a well zone. Unit has 36 well zones (shading).
Figure 7-9. 4-Well Design for Production Unit with Low Connectivity
Draft Technical Report
153
Revised Draft - November 26, 2012
-------
Production Unit with 7-spot pattern, 126 injection wells (•) and 36 production wells ({°}).
Unit has 16 well zones (shading) and an average zone contains 10 wells.
Figure 7-10. 10-Well Design for Production Unit with Good Connectivity
Draft Technical Report
154
Revised Draft - November 26, 2012
-------
• ••••••
• ••••••
• •••••
• ••••••
• ••••••
• •••••
• •••••
•
:
Production Unit with 7-spot pattern, 126 injection wells (•) and 36 production wells ({°}).
Unit has 9 well zones (shading) and an average zone contains 18 wells.
Figure 7-11. 18-Well Design for Production Unit with High Connectivity
Draft Technical Report
155
Revised Draft - November 26, 2012
-------
Table 7-9. Ratio of Monitor Wells to Production Wells
Site
Highland Uranium
Project
Irigary
Christensen Ranch
Crow Butte
Totals
Averages
Production
Unit
A-Wellfield
Units 1 to 9
Units 2 to 5
Unitl
Actual Number of
Ore Zone Monitor
Wells
5
45
78
12
140
9.3
Number of
Extraction
Wells*
38
453
663
44.2%
52
1,206
80.4
Number of
Injection
Wells*
48
565
838
55.8%
64
1,515
101
Total
Number of
Wells
86
1,018
1,501
116
2,721
181.4
: Other sites estimated using Christensen Ranch proportion of injection/extraction wells
Table 7-10. Number of Monitoring Wells Required for Five Design Options for a
Production Unit with 181 Wells in 5-spot Pattern (Posterior Probability=90%)
Posterior Probability=90%
Design Option
2-Well Design
5 -Well Design
7-Well Design
11 -Well Design
20-Well Design
Number
of Well
Zones
81
36
25
16
9
Percent of Well Zones Demonstrating Compliance (P)
P=70
6
5
5
5
4
75
7
7
6
5
4
80 85 90 91 92 93 94 95 96 97 98 99
9 12 17 19 21 25 29 29 34 42 55 72
8 10 15 15 18 18 18 24 24 24 32 32
7 10 13 13 13 17 17 17 17 23 23 23
6 8 11 11 11 11 15 15 15 15 15 15
66
Note: All clusters must be monitored.
Table 7-11. Number of Monitoring Wells Required for Five Design Options for a
Production Unit with 181 Wells in 5-spot Pattern (Posterior Probability=95%)
Posterior Probability=95%
Design Option
2-Well Design
5 -Well Design
7-Well Design
11 -Well Design
20-Well Design
Number
of Well
Zones
81
36
25
16
9
Percent of Well Zones Demonstrating Compliance (P)
P=70
7
7
6
6
5
75
9
8
7
6
5
80 85 90 91 92 93 94 95 96 97 98 99
11 15 21 24 27 30 35 35 41 50 62 76
10 13 18 18 22 22 22 27 27 27 34 34
9 12 15 15 15 19 19 19 19 24 24 24
8 10 12 12 12 12
77
Note: All clusters must be monitored.
Draft Technical Report
156
Revised Draft - November 26, 2012
-------
Table 7-12. Number of Monitoring Wells Required for Four Design Options for a
Production Unit with 162 Wells in 7-spot Pattern (Posterior Probability=90%)
Posterior Probability=90%
Design Option
2-Well Design
4-Well Design
10-Well Design
18-Well Design
Number
of Well
Zones
72
36
16
9
Percent of Well Zones Demonstrating Compliance (P)
P=70
5
5
5
4
75 80 85 90 91 92 93 94 95 96 97 98 99
7 9 12 16 19 21 25 25 30 37 37 48 63
7 8 10 15 15 18 18 18 24 24 24 32 32
5 6 8 11 11 11 11 15 15 15 15 15 15
466
Note: All clusters must be monitored.
Table 7-13. Number of Monitoring Wells Required for Four Design Options for a
Production Unit with 162 Wells in 7-spot Pattern (Posterior Probability=95%)
Posterior Probability=95%
Design Option
2-Well Design
4-Well Design
10-Well Design
18-Well Design
Number
of Well
Zones
72
36
16
9
Percent of Well Zones Demonstrating Compliance (P)
P=70
7
7
6
5
75 80 85 90 91 92 93 94 95 96 97 98 99
9 11 15 21 23 26 31 31 36 44 44 54 67
8 10 13 18 18 22 22 22 27 27 27 34 34
6 8 10 12 12 12 12
577
Note: All clusters must be monitored.
Draft Technical Report
157
Revised Draft - November 26, 2012
-------
Table 7-14. Number of Monitoring Wells Required for Posterior Probability of 90% that
at Least P% of the Well Zones Demonstrate Compliance
Number of
Well Zones
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
25
30
35
40
45
50
55
60
65
70
75
80
85
90
95
100
110
120
130
140
150
160
170
180
190
200
Percent
P=70
3
4
o
5
4
4
4
4
4
5
4
5
5
4
5
5
4
5
5
5
5
5
5
6
5
6
5
6
6
6
6
6
6
6
6
6
6
6
6
6
6
6
6
75
3
4
5
4
4
5
5
4
5
5
6
5
5
6
6
5
6
6
7
6
7
7
7
7
7
7
7
7
7
7
7
7
7
7
7
7
7
7
7
7
7
7
80
3
4
5
5
6
5
5
6
6
7
6
6
7
7
8
6
7
7
8
8
8
8
8
8
8
9
9
9
9
9
9
9
9
9
9
9
9
9
9
9
9
9
85
-
-
5
5
6
7
7
8
9
7
8
8
9
9
10
8
10
10
10
10
12
11
11
11
12
12
12
12
12
12
12
12
13
12
13
12
13
13
13
13
13
13
of Well
90
-
-
-
-
-
7
7
8
9
9
10
11
11
12
13
10
13
12
15
14
16
15
17
16
17
16
18
17
18
17
18
18
18
18
18
19
19
19
19
19
19
19
Zones Demonstrating
91 92
-
-
-
-
-
9 9
10 10
8 11
9 9
9 9
10 10
11 11
11 11
12 12
13 13
14 14
13 13
16 16
15 18
17 17
16 19
18 18
19 19
18 21
20 20
19 21
20 20
19 21
20 23
19 21
20 23
19 21
21 24
21 23
21 23
21 23
21 23
21 25
21 24
21 24
21 24
21 24
93 94
-
-
-
-
-
9 9
10 10
11 11
12 12
13 13
10 14
11 15
11 11
12 12
13 13
14 14
17 17
16 20
18 18
21 21
19 24
21 21
23 23
21 26
23 28
25 25
23 27
25 29
26 26
24 28
26 29
27 27
26 30
26 29
26 31
28 30
27 30
27 32
28 31
28 33
27 32
29 31
Compliance (P)
95
-
-
-
-
-
9
10
11
12
13
14
15
16
17
18
14
17
20
24
21
24
26
29
26
28
30
32
29
30
32
34
31
34
33
35
34
36
35
37
36
38
37
96 97
-
-
-
-
-
9 9
10 10
11 11
12 12
13 13
14 14
15 15
16 16
17 17
18 18
18 18
17 23
20 27
24 24
27 27
31 31
26 34
29 37
32 41
34 44
37 37
32 40
34 42
36 45
39 48
41 50
36 43
40 47
43 52
40 56
44 51
41 54
44 58
47 53
44 56
46 60
44 55
98
-
-
-
-
-
9
10
11
12
13
14
15
16
17
18
18
23
27
32
36
41
34
37
41
44
48
51
55
58
61
65
53
58
64
69
75
65
69
74
78
82
73
99
-
-
-
-
-
9
10
11
12
13
14
15
16
17
18
18
23
27
32
36
41
45
50
54
59
63
68
72
77
81
86
68
75
82
89
96
102
109
116
123
130
107
Draft Technical Report
158
Revised Draft - November 26, 2012
-------
Table 7-14. Number of Monitoring Wells Required for Posterior Probability of 90% that
at Least P% of the Well Zones Demonstrate Compliance
Number of
Well Zones
225
250
300
350
400
450
500
600
700
800
900
1000
P=70
6
6
6
6
6
6
6
6
6
6
6
6
75
7
7
7
7
7
7
7
7
7
7
7
7
Percent
80 85
9 13
9 13
9 13
10 13
10 13
10 13
10 13
10 13
10 13
10 13
10 13
10 13
of Well
90
20
20
20
20
20
20
20
21
21
21
21
21
Zones Demonstrating
91 92
22 24
22 25
22 25
23 25
23 25
23 26
23 26
23 26
23 26
23 26
23 26
23 26
93 94
29 33
29 32
30 33
29 33
30 34
30 34
30 34
30 35
31 35
31 35
01 o e
31 35
01 o e
31 35
Compliance
95
38
39
39
41
40
41
41
41
42
42
42
43
96
45
46
47
48
49
50
50
51
52
52
53
53
(P)
97
62
61
60
65
64
67
66
67
68
69
70
70
98
82
79
83
86
89
91
93
96
98
100
101
102
99
120
133
131
152
147
165
158
167
174
179
184
188
Note: All clusters must be monitored.
Table 7-15. Number of Monitoring Wells Required for Posterior Probability of 95% that
at Least P% of the Well Zones Demonstrate Compliance
Number of
Well Zones
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
25
30
35
40
45
50
55
Percent of Well Zones
P=70
4
4
4
4
5
4
5
5
6
5
6
6
5
6
6
6
6
6
7
7
7
7
7
75
4
4
5
4
5
6
6
5
6
7
7
6
7
7
8
7
7
8
8
8
8
9
9
80
4
4
5
6
7
6
6
7
8
8
7
8
8
9
9
8
9
9
10
10
10
10
11
85
-
-
5
6
7
8
8
9
10
8
9
10
10
11
11
10
12
13
13
13
14
14
14
90 91
-
-
-
-
-
8
8
9 9
10 10
11 11
12 12
12 12
13 13
14 14
15 15
12 15
15 15
15 18
18 18
17 20
19 19
19 22
20 24
Demonstrating Compliance
92
-
-
-
-
-
-
-
-
10
11
12
12
13
14
15
15
15
18
22
20
23
22
24
93 94
-
-
-
-
-
-
-
-
-
-
12
12
13 13
14 14
15 15
15 15
19 19
18 23
22 22
25 25
23 28
26 26
28 28
95
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
15
19
23
27
25
28
31
34
(P)
96 97
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
19 19
19 24
23 29
27 27
31 31
35 35
31 39
34 43
98
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
19
24
29
34
38
43
39
43
99
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
19
24
29
34
38
43
48
53
Draft Technical Report
159
Revised Draft - November 26, 2012
-------
Table 7-15. Number of Monitoring Wells Required for Posterior Probability of 95% that
at Least P% of the Well Zones Demonstrate Compliance
Number of
Well Zones
60
65
70
75
80
85
90
95
100
110
120
130
140
150
160
170
180
190
200
225
250
300
350
400
450
500
600
700
800
900
1000
Percent of Well
P=70
7
7
7
7
7
7
7
7
7
7
7
7
7
8
8
8
8
8
8
8
8
8
8
8
8
8
8
8
8
8
8
75
9
9
9
9
9
9
9
9
9
9
9
9
9
9
9
10
9
10
9
10
10
10
10
10
10
10
10
10
10
10
10
80
11
11
11
11
11
11
11
11
12
12
12
12
12
12
12
12
12
12
12
12
12
12
12
13
13
13
13
13
13
13
13
85 90
14 20
15 21
15 21
15 22
15 21
16 23
16 22
16 23
15 22
16 23
16 23
16 23
16 24
17 24
16 24
17 24
16 25
17 25
17 25
17 26
17 25
17 26
17 26
17 26
17 26
17 27
17 27
18 27
18 27
18 27
18 27
Zones
91
22
24
23
25
24
25
24
26
24
27
27
27
27
27
27
27
27
27
27
28
29
29
29
29
30
30
30
30
30
30
30
Demonstrating Compliance
92
26
24
26
25
27
28
27
28
27
30
30
30
29
29
31
31
31
31
31
31
32
32
33
33
33
33
34
34
34
34
34
93
26
28
31
28
30
32
30
32
34
33
33
32
35
34
34
36
35
35
37
37
37
38
38
39
38
39
39
40
40
40
40
94
31
33
31
33
35
32
34
36
34
37
36
39
38
37
40
39
41
40
40
42
41
42
43
43
44
44
45
45
46
46
46
95
31
33
36
39
35
37
40
42
38
42
41
44
42
46
44
47
45
48
46
48
50
50
52
51
53
53
53
54
55
55
55
(P)
96 97
37 46
41 50
44 44
39 47
41 50
44 53
47 56
49 59
44 52
49 57
53 62
50 68
54 62
51 67
54 71
58 66
55 70
58 74
55 68
57 77
58 77
60 76
62 82
63 81
64 85
65 84
66 86
67 87
68 89
68 90
69 90
98 99
46 57
50 62
54 67
58 72
62 76
66 81
70 86
74 91
63 78
69 85
75 93
82 101
88 109
78 116
84 124
89 132
94 140
99 147
89 126
100 142
97 157
103 157
108 184
112 179
115 202
118 195
122 208
125 217
128 225
130 232
131 237
Note: All clusters must be monitored.
7.2.3 Determining Connectivity of the Well field
Connectivity of the wellfield may be measured using both physical and statistical methods. An
example of a physical measure of connectivity is the hydraulic conductivity of the ground water.
Connectivity may also be demonstrated based on an analysis of the spatial correlation between
measurements in nearby wells. Spatial autocorrelation is a generalization of one-dimensional
temporal autocorrelation. Spatial correlation may be defined in two, three or more dimensions. In
this application two-dimensional spatial autocorrelation is of interest. In two or more dimensions
autocorrelation is multi-directional.
Draft Technical Report
160
Revised Draft - November 26, 2012
-------
Two- and three-dimensional data with spatial correlation are best analyzed using geostatistical
methods. Geostatistics is capable of using the information revealed by a correlation analysis of
the data to estimate concentrations at unsampled locations and in areas where data are sparse.
Although application of these methods often requires specialized software and skilled
practitioners, operators are encouraged to perform geostatistical analysis to obtain a better
understanding of the site. Geostatistical analysis may reveal anomalous baseline conditions in
certain areas of the site or areas where remediation has not been successful. A full discussion of
geostatistics is beyond the scope of this document. Some examples of the application of
geostatistics in the analysis of ground water contamination can be found in Chapter 6 of EPA
2000b.
Moran's / statistic is one measure of spatial association that does not require specialized
software. This statistic is a global measure of spatial autocorrelation characterized by a
correlation among nearby wells. In addition to the data from each well, Moran's / statistic
requires matrix of spatial weights. In the most elementary case, the weight matrix may be
derived from a distance matrix. The distance matrix has elements that represent the "distance"
between well / and welly. The distance matrix may be expressed in physical units of length, or
may contain other measures of distance, such as the travel time between wells.
The weight matrix for the / statistic has elements equal to the inverse of the elements of the
distance matrix. The weight matrix measures the "nearness" of each well to the other wells. The
weighting entries for pairs of wells that are close together are higher than for pairs of wells that
are far apart. The weight matrix has off-diagonal entries witj equal to 1 divided by the distance
between well /' and welly. The diagonal entries of the weight matrix are set to 0. This is one of
several ways to calculate an inverse distance matrix. For example, the inverse distance matrix
may have off-diagonal entries equal to 1 divided by 1 plus the distance between well /' and welly.
Moran's /is calculated using the formula:
/= N
where TV is the number of spatial units indexed by i,j=l, ..., N; Xis the analyte of interest; X is
the mean of X; and wtj is an element of the spatial weight matrix.
The expected value of Moran's / statistic under a hypothesis of no spatial autocorrelation is
£(/) = - \/(N - l). Positive (negative) values of/indicate positive (negative) spatial
autocorrelation. Values of/range from -1, indicating perfect dispersion, to +1, indicating perfect
correlation between the wells. A zero value indicates a random spatial pattern. For statistical
hypothesis testing, Moran's /values can be transformed to normal scores in which values greater
than 1.96 (or smaller than -1.96) indicate spatial autocorrelation that is significant at the 5%
level.
Draft Technical Report 161 Revised Draft - November 26, 2012
-------
7.3 Hypothesis Testing and Data Quality Objectives
Hypothesis testing is a statistical tool for deciding when the ground water has reached steady
state and for comparing post-restoration conditions with baseline conditions. The hypothesis tests
are conducted for individual wells and, when wells exhibit homogeneous dynamics, for all wells
combined.
The first step in developing a hypothesis test is to transform the problem into statistical
terminology by formulating a null hypothesis and an alternative hypothesis. These hypotheses
form the two alternative decisions that the hypothesis test will evaluate. When a well is
compared with the baseline, the unknown parameter of interest (<§) is the amount by which the
post-restoration distribution exceeds the baseline distribution (See Box 7-1). Delta (S) is an
unknown value, and statistical tests may be used to evaluate hypotheses relating to its possible
values. A hypothesis test is designed to reject or not reject hypotheses about S based on test
statistics computed from the sample data.
At its core, this is another example of the "How clean is clean?" problem. The action level for
baseline comparisons is the largest difference in the two distributions that is acceptable to the
decision maker. In this report, the action level for this difference is defined as a substantial
difference (A), which may be zero or a positive value based on the risk assessment, an applicable
regulation, a screening level, or guidance.
This document does not establish a specific value for a substantial difference A, as the value will
vary from parameter to parameter and from site to site. Therefore, specific values for A should be
considered on a case-by-case basis. In many cases, the minimum feasible value of A is
determined by the usual variability in that parameter during the pre-leaching phase (Phase 1 in
Figure 3-1). Appendix A to EPA 2002a discusses further the selection of a value for A. The
determination of A for each parameter of interest should be considered during the development
of a quality assurance project plan as part of the planning process for the site evaluation.
Hypothesis testing is a quantitative method to determine whether a specific statement concerning
the unknown difference 5 (a statement known as the "null hypothesis") can be rejected based on
the data at hand. Decisions concerning the true value of 5 (e.g., is 5 > 0?) reduce to a choice of
"yes" or "no." When viewed in this way, two types of incorrect decisions, or decision errors,
may occur:
• Incorrectly deciding the answer is "yes" when the true answer is "no"
• Incorrectly deciding the answer is "no" when the true answer is "yes"
While the possibility of decision errors can never be totally eliminated, it can be controlled to
acceptable levels. To control decision errors, it is necessary to control the uncertainty in the
estimate of 5. Uncertainty arises from three sources:
• Sampling error
• Measurement error
• Natural variability
Draft Technical Report 162 Revised Draft - November 26, 2012
-------
The decisionmaker has some control over the first two sources of uncertainty. For example, a
larger number of samples may lead to fewer decision errors because the probability of a decision
error decreases as the number of samples increases. Use of more precise measurement techniques
or duplicate measurements can reduce measurement error, thus minimizing the likelihood of a
decision error. The third source of uncertainty is more difficult to control. Natural variability
arises from the uneven distribution of chemical concentrations and conditions at the site.
Natural variability includes both spatial and temporal variability. When measurements are made
in different wells over a relatively short period of time, spatial variability is measured by the
standard deviation (a) of the measurements around their mean value. A large value of o indicates
that a large number of measurements will be needed to achieve a desired limit on decision errors.
Baseline samples may have a different variability than post-restoration samples. As post-
restoration variability is usually higher than in the baseline, post-restoration data ideally would
be used to estimate o.
Temporal variability measures the variability of concentration in a well over time. Short-term
variability may be random or have a seasonality pattern. Long-term temporal variability may
appear as a trend. In the case of a trend, temporal variability is defined in terms of the variation
of the measurements from the trendline.
It is advisable to overestimate o rather than underestimate the true variability. A very crude
approximation for o may be made by dividing the anticipated range (maximum minus minimum)
by 6 (EPA 2002a, Section 3.1). It is important that overly optimistic estimates for o be avoided,
because this may result in a sample size that fails to generate sufficient data to distiguish between
the null and alternative hypotheses. In statistical terms, the test will lack sufficient power if the
sample size is too small. The power of a statistical hypothesis test is discussed in the following
section.
The minimum detectable difference (MDD) for a statistical test indicates that differences smaller
than the MDD cannot be detected reliably. If the test is used to decide if post-restoration
concentrations exceed the baseline concentrations by more than A, it is necessary to ensure that
the MDD for the test is less than A. In the planning stage, this requirement is met by designing a
sampling plan with sufficient power to detect differences as small as A (MDD < A). If data were
collected without the benefit of a sampling plan, retrospective calculation of the power of the test
may be necessary before making a decision.
In the planning stage, the absolute size of the MDD is of less importance than the ratio of the
MDD to the natural variability of the post-restoration concentrations. This ratio is termed the
relative difference and is defined as MDD/o, where o is an estimate of the standard deviation of
the post-restoration distribution. The relative difference expresses the power of resolution of the
statistical test (MDD) in units of uncertainty (o). Relative differences much less than 1 standard
deviation (MDD/o « 1) are more difficult to resolve unless a larger number of measurements
are available. Relative differences of more than 3 standard deviations (MDD/o > 3) are easier to
resolve.
Draft Technical Report 163 Revised Draft - November 26, 2012
-------
7.3.1 Decision Errors and Confidence Levels
A key step in developing a sampling and analysis plan is to establish the level of precision
required of the data used for decision making. These requirements will determine the required
sample size. An increased number of samples generally increases the level of precision. Due to
the uncertainties that result from sampling variation, decisions will be subject to errors. There are
two ways to err when analyzing data (Table 7-16):
• Type I Error: Based on the observed data, the test may reject the null hypothesis when, in
fact, the null hypothesis is true (a false positive). This is a Type I error. The probability of
making a Type I error is a (alpha).
• Type IIError: On the other hand, the test may fail to reject the null hypothesis when the
null hypothesis is, in fact, false (a false negative). This is a Type II error. The probability
of making a Type II error is P (beta).
Table 7-16. Hypothesis Testing: Type I and Type II Errors
Decision Based on
Sample Data
H0 is not rejected
H0 is rejected
Actual Site Condition
H0 Is True
Correct Decision: (1 - a)
Type I Error: False
Positive (a)
H0 Is Not True
Type II Error: False
Negative (ft)
Correct Decision: (1 - (3)
Note: H is the null hypothesis.
The acceptable level of decision error associated with hypothesis testing is defined by two key
parameters: confidence level and power. These parameters are closely related to the two error
probabilities, a and p.
• Confidence level: 100(1 - a)%. As the confidence level is lowered (or
alternatively, as a is increased), the likelihood of committing a Type I error
increases.
• Power: 100(1 - fi)%. As the power is lowered (i.e., as P is increased), the
likelihood of committing a Type II error increases.
The selection of appropriate levels for decision errors and the resulting number of samples is a
critical component of the DQO process that should concern all stakeholders.
Because there is an inherent tradeoff between the probability of committing a Type I or Type II
error, a simultaneous reduction in both types can occur only by increasing the number of
samples. If the probability of committing a false positive is reduced by increasing the level of
confidence of the test (in other words, by decreasing a), the probability of committing a false
negative is increased, because the power of the test is reduced (increasing P).
When the site is sampled for a number of species, the selection of appropriate DQOs for each
contaminant will be influenced by the relative health risks and costs of control for each species.
If a single contaminant is the major focus of concern, the DQOs (a and P) may be based on this
Draft Technical Report
164
Revised Draft - November 26, 2012
-------
species. If more than one species is a matter of concern, the Bonferroni correction19 is a simple
approach for addressing the problem. If the species are of equal concern, the nominal
significance level for each test (a) is divided by the number of contaminants that are to be tested.
Similarly, when the species have different levels of concern, adjustments may be made using a
different a for each species. However, the sample size calculations (described below) show that
this reduction in the significance level requires a major increase in the amount of data to be
collected. The issue of multiple comparisons is beyond the intended scope of this document. A
complete discussion of Bonferroni confidence intervals and newer alternative approaches to the
multiple-comparison problem is presented in Bickel and Doksum (2006). In terms of not
exceeding the regulatory limits, we believe the confidence level for decisions involving the listed
radionuclides should be equivalent. The limits for other analytes may be less stringent but still
consistent with the value of the parameter in making a safety case.
7.3.2 Hypothesis Tests for Comparisons with Baseline
Statistical hypothesis tests are used for comparing post-restoration conditions with baseline
conditions and for demonstrating stability of the site after restoration. Three statistical methods
are presented for the comparison with baseline conditions. The two-sample t-test and prediction
limits for a future mean are two parametric methods used in the comparison with baseline
conditions. Prediction limits (PLs) are designed to provide an upper bound for the mean of a
future sample with a specified probability equal to (1 - a), known as the confidence level of the
PL. It represents the chance that the PL will contain the mean of a future (post-operations)
sample from the monitoring wells. The nonparametric WRS test is also used to compare post-
restoration well conditions with baseline values. This test compares the relative ranks of the two
data sets. The two-sample Mest and the WRS test are recommended for comparing baseline and
post-remedial wells (EPA 2006a). The RCRA Unified Guidance (EPA 2009) recommends the
PL method.
The three different statistical tests for comparisons with baseline conditions are used in Phase 4
to determine if the restoration goals have been met. A choice of one of the three different
statistical tests should be made before post-restoration data are collected. The selected test
procedure is then applied for all wells. The prior selection of a single testing approach is
necessary for comparability of results across wells and to avoid the possibility of running all
three tests, then selecting the test with most favorable results.
Two forms of hypothesis tests are used for comparisons with baseline conditions. The null
hypothesis in the first form of the test (Test Form 1) states that there is no statistically significant
difference between the means of the baseline and post-restoration concentration distributions.
The null hypothesis in the second form of the test (Test Form 2) states that the post-restoration
mean exceeds the baseline mean by more than a substantial difference A. Either test form may be
used with the two-sample Mest and the WRS test. Test Form 1 is used for PLs.
19 The Bonferroni correction is a statistical method used to address the problem of multiple comparisons. It
helps control the probability of Type I errors (i.e., false positives).
Draft Technical Report 165 Revised Draft - November 26, 2012
-------
Test Form 1
The null hypothesis for Test Form 1 is formulated for the express purpose of being rejected when
restoration has not been successful. If the null hypothesis is rejected, the alternative hypothesis is
accepted, indicating that baseline conditions have not been achieved.
• The null hypothesis (Ho): The mean of the post-restoration distribution does not exceed
the baseline mean. Symbolically, the null hypothesis for Test Form 1 is HQ: 5 < 0.
• The alternative hypothesis (HA): The post-restoration distribution mean exceeds the
baseline mean: (HA: 5 > 0).
When using Test Form 1, the null hypothesis is presumed true until it is rejected. A problem may
arise in the use of Test Form 1 when insufficient data are available for the decision. If
insufficient data are available, Test Form 1 may lead to inconclusive results due to a lack of
power. Figure 7-12 illustrates an example of a test performance plot for Test Form 1. At the
origin of the plot, the true difference between the means of the two distributions is zero. Positive
values of the difference are plotted on the horizontal axis to the right of the origin, negative
values to the left. The vertical axis shows the probability of rejecting the null hypothesis, which
for this test form is the probability of deciding the post-restoration mean concentration exceeds
the mean baseline concentration. This probability ranges from 0 to 1.0 (0% to 100%).
The gray region for the test extends from 0 up to the MDD. It is necessary to specify a gray
region for the test, because the decision may be "too close to call" due to uncertainty. This may
occur when the difference in means is small compared to the MDD for the test. To the left or
right of the gray region, the test outcome is easy to determine. In the gray region, the statistical
test has difficulty deciding between the two alternatives. To the left of the gray region, the test
performance curve is no greater than a. In the gray region, the test performance curve increases
as the difference between the means increases. The number of samples and the standard
deviation, o, determine the rate of increase. The right edge of the gray region is located at the
MDD. The MDD measures the width of the gray region for the test. When the difference
between the means is equal to the MDD, the probability of deciding that the post-restoration
mean exceeds the baseline mean is equal to the power of the test (1 ~P).
Figure 7-12 also shows a hypothetical value of a substantial difference for this analyte of
A = 100. This value is used in the DQO process as an upper limit for the width of the gray region
(MDD). In this example, an MDD less than A was selected for the test. If the MDD is selected to
be smaller than A, then differences from baseline smaller than A can be detected by the test with
a high probability.
Draft Technical Report 166 Revised Draft - November 26, 2012
-------
Test Performance Plot: Test Form 1
Ho: 6 < 0 versus HA: 5 > 0
100%
E
ro
A.
• The alternative hypothesis (HA): The post-restoration distribution mean does not exceed
the baseline mean by more than A (HA: 5 < A).
In Test Form 2, the hypothesis test is structured so that the post-restoration data must provide
evidence that the site is within acceptable limits. The null hypothesis is assumed to be true unless
the statistical test indicates that it should be rejected in favor of the alternative. Figure 7-13
illustrates an example of a test performance plot for Test Form 2. The horizontal and vertical
axes for this plot are the same as in Figure 7-12. The gray region for Test Form 2 again has a
width equal to the MDD. To the left of the gray region, the test performance curve is no greater
than p. In the gray region, the test performance curve increases as the difference between the
means increases. The right edge of the gray region is located at A, the value selected as a
substantial difference. When the difference between the means is equal to A, the probability of
deciding that the post-restoration mean exceeds the baseline mean is equal to (1 - a).
Draft Technical Report
167
Revised Draft - November 26, 2012
-------
An MDD less than A was selected for the test. If the MDD is selected to be smaller than A, then
the test will have greater power to accept sites when the difference in means is less than A.
Test Performance Plot: Test Form 2
Ho: 6 > A versus HA: 6 < A
100%
11
= -a
O c
— QJ
1° I
£ I
o 1
>• rc
•= a
5 S
s s
Q. LU
60%
40%
20%
-40 -20 0 20 40 60 80 100 120 140
6 = Postrestoration minus Baseline Concentration
Figure 7-13. Test Performance Plot with Parameter Definitions for Test Form 2
7.3.3 Selecting a Test Form
When comparing Test Forms 1 and 2, it is important to distinguish between the selection of the
null hypothesis, which is a burden-of-proof issue, and the selection of the substantial difference,
which involves determination of an action level.
Test Form 1 uses a conservative level of 0 as the maximum acceptable difference, but relaxes the
burden of proof by selecting a null hypothesis that the post-restoration mean is not statistically
different from the baseline mean. Test Form 2 requires a stricter burden of proof, but relaxes the
action level from 0 to A. See Box 7.2 for more information on the parameters of Test Form 1 and
Test Form 2. EPA2002a and MARSSEVI (EPA 2000a) include further discussion of how to
choose between Test Forms 1 and 2 (MARSSEVI uses the terms Scenario A and Scenario B for
Test Forms 2 and 1, respectively), and provide additional guidance for setting up the hypotheses.
Regardless of the choice of hypothesis for the comparison with baseline conditions, an incorrect
conclusion could be drawn from the data analysis using either form of test. To account for this
inherent uncertainty, one should specify the limits on the decision errors. This task was described
in Section 7.3.1.
The selection of a null hypothesis depends on what the "working assumption" is for each
monitoring phase and, perhaps more generally, what has already occurred. For post-restoration,
one "assumes the worst," i.e., that post-restoration values exceed the baseline by more than A.
Compliance can be demonstrated only by collecting sufficient data to reject the null hypothesis
Draft Technical Report
168
Revised Draft - November 26, 2012
-------
(i.e., "proving" the alternative). For this phase, a regulator would be primarily concerned with
the occurrence of Type I errors (showing compliance when not justified). However, when using a
trend test to demonstrate that the site is stable, the null hypothesis would be a statement of no
trend. The null hypothesis would stand until sufficient data show evidence of a trend.
A Type 2 error occurs if the null hypothesis is accepted when it is not true. With Test Form 1, a
Type 1 error occurs when the site is incorrectly determined to require further restoration. A
Type 2 error means that the site was incorrectly determined to be in compliance. From a human
health perspective, a Type 2 error is more serious than a Type 1 error. Hence, it is reasonable that
the Type 2 error rate (P) should be smaller than the Type 1 error rate (a). Using Test Form 1, the
regulator would be concerned with Type 2 errors, and the specification of MDDs and minimum
sample sizes will be of particular importance to the regulator.20
With Test Form 2, a Type 1 error occurs when the site is incorrectly determined to be in
compliance. A Type 2 error means that the site was incorrectly determined to require further
restoration. From a human health perspective, a Type 1 error is more serious than a Type 2 error.
In this case, it is reasonable that the Type 1 error rate (a) should be smaller than the Type 2 error
rate (P). Once a is selected, a higher value of P will reduce the required number of samples, but
there will be a greater likelihood that the site is incorrectly determined to be out of compliance.
In this case, the site operator faces a trade-off and may select to reduce the value of P (at the
expense of a greater number of samples) and increase the power of the test in order to avoid the
possibility of Type 2 errors.
For the purposes of this report, the minimum recommended performance measures are:
Test Form 1:
Confidence level at least 80% (a < 0.20) and power at least 90% (p <0.10)
Test Form 2:
Confidence level at least 90% (a < 0.10) and power at least 80% (p < 0.20).
(EPA2002a, Section 3.2)
Box 7-2 describes these performance standards in more detail.
Documents such as EPA (2009) describe processes involving several phases, and the null hypothesis
depends on the phase of the process and/or what may have occurred previously. For some stages, the null hypothesis
would be a statement that "all is well" (e.g., there is no trend for a particular contaminant at a particular well
monitoring location). For other stages, the null hypothesis is just the opposite (e.g., the site is out of compliance with
respect to a particular contaminant). For the former case, rejection of the null hypothesis in effect "proves" that the
site is not stable, and regulators would be primarily concerned with the occurrence of what statisticians refer to as
Type 2 errors (we are unable to detect a worrisome trend when such a trend exists). For the latter, the primary
concern would be the occurrence of a Type I error (falsely concluding that the site is in compliance when it is not).
Draft Technical Report 169 Revised Draft - November 26, 2012
-------
Box 7-1. Definitions
3 (delta): The true difference between the post-restoration distribution and the baseline
distribution of parameter X. Delta is an unknown value that describes the true state of nature.
Hypotheses about its value are evaluated using statistical hypothesis tests. In principle, we can
select any specific value for S and then test if this difference is statistically significant or not with
a given confidence and power.
A (a substantial difference): A difference between the two distributions that is sufficiently large
to warrant additional interest based on health or ecological information. A is the investigation
level. If 5 exceeds A, the difference in concentrations is judged to be sufficiently large to be of
concern for the purpose of the analysis. A hypothesis test uses baseline and post-restoration
measurements to determine if 5 exceeds A.
MDD (minimum detectable difference): The smallest difference that the statistical test can
resolve. The MDD depends on sample-to-sample variability, the number of samples, and the
power of the statistical test. The MDD is a property of the survey design.
Box 7-2. Interpretation of the Statistical Measures for Test Forms 1 and 2
Test Form 1
Confidence level = 80%: On average, in 80 out of 100 cases, post-restoration concentrations are
correctly identified as not exceeding baseline concentrations by more than A, while in 20 out of
100 cases, post-restoration concentrations will be incorrectly identified as exceeding baseline
concentrations by more than A when, in fact, they do not. For a confidence level of 80%, choose
a=0.20.
Power = 90%: On average, in 90 out of 100 cases, post-restoration concentrations will be
correctly identified as exceeding baseline concentrations by more than A, while in 10 out of 100
cases, post-restoration concentrations will be incorrectly identified as not exceeding baseline
concentrations by more than A when, in fact, they do. For power of 90%, choose |3=0.10.
Test Form 2
Confidence level = 90%: On average, in 90 out of 100 cases, post-restoration concentrations are
correctly identified as exceeding baseline concentrations by more than A, while in 10 out of 100
cases, post-restoration concentrations will be incorrectly identified as not exceeding baseline
concentrations by more than A when, in fact, they do. For a confidence level of 90%, choose
a=0.10.
Power = 80%: On average, in 80 out of 100 cases, post-restoration concentrations will be
correctly identified as not exceeding baseline concentrations by more than A, while in 20 out of
100 cases, post-restoration concentrations will be incorrectly identified as exceeding baseline
concentrations by more than A when, in fact, they do not. For power of 80%, choose |3=0.20.
7.3.4 Hypothesis Tests for Detecting Trends
Two statistical tests for trends are used for stability monitoring in Phase 5, a parametric trend test
based on the linear regression model and the nonparametric Mann-Kendall trend test. A choice
of one these statistical tests should be made before stability monitoring is begun. The selected
Draft Technical Report 170 Revised Draft - November 26, 2012
-------
test procedure is then applied for all monitored wells. Selection of a single testing approach is
necessary for comparability of results across wells and to avoid the possibility of running both
tests, then selecting the test with most favorable results.
Linear regression and the nonparametric Mann-Kendall trend test are recommended for trend
detection in EPA 2006a and EPA 2009. The linear regression trend test relies on a variety of
assumptions (e.g., normality) that must be verified. The Mann-Kendall trend test may be used
with any series of four or more independent samples to test for trends in well parameters. Trend
tests are used in Phase 1 to check for unexpected trends in baseline samples, and in Phase 5 to
establish long-term stability.
Test Form 1 is used for the regression t-test and the Mann-Kendall trend test. In regression, a
trend is measured by the slope of the regression line 9 (theta). Here 9 represents the true (and
unknowable) value of the slope of the trend line. In the Mann-Kendall test, a trend is charactered
by the parameter T (tau), which is a nonparametric measure of the correlation of the sample
values with time. In each case, hypothesis tests are used to determine if there is a significant
trend over time. The null hypothesis for both tests is that there is no trend. The time series is
analyzed for evidence of a significant trend. If such evidence is found, then the null hypothesis is
rejected.
Testing for a trend:
• The null hypothesis (Ho): There is no significant trend in the series. Symbolically, the
null hypothesis for the trend test is HQ: 9=0.
• The alternative hypothesis (HA): There is a significant trend in the series. (HA: 9 ^ 0).
The null and alternative hypotheses for the Mann-Kendall test are similar, with 9 being replaced
by T.
Unlike the tests for a comparison with baseline conditions, the trend tests require only one DQO
parameter, the Type 1 error rate a. This parameter provides a control on the frequency of false
positives, i.e., incorrectly deciding there is a trend when, in fact, the series is stationary. The
power (1-P) of the trend tests to detect a trend depends on the variability of the series, the type of
trend, duration of the sampling program and the frequency of sampling. The power of these tests
may be estimated using Monte Carlo simulation. The power of the rank-based Mann-Kendal test
and the linear regression t-test for slope are compared by Yue and Pilon (2004), who report that
the power of the t-test is slightly higher than that of the Mann-Kendall test for normally
distributed data, and for nonnormally distributed series, such as time series with the Weibull,
Gumbel, Pearson Type III, or extreme value Type II distributions, the power of the Mann-
Kendall test is higher than that of the t-test. The t-test usually performs better with a distribution
that is relatively symmetric. In contrast, the Mann-Kendall is a more powerful test for data sets
that present skewness. Attachment G contains a detailed discussion of the simulations performed
for this study to determine a sufficient number of samples to provide adequate power to detect
trends.
Draft Technical Report 171 Revised Draft - November 26, 2012
-------
Adopting hypothesis tests and a DQO approach described in EPA QA/G9S (EPA 2006a,
Section 3.4) can help control the probability of making decision errors. However, incorrect use
of hypothesis tests can lead to incorrect decisions. Each type of hypothesis test is based on a set
of assumptions that should be verified to confirm proper use of the test. The tests recommended
in this document for verifying stability and determining when the site has met the remedial goals
include both parametric and nonparametric tests. Nonparametric tests generally have fewer
assumptions to verify. The following section provides additional information on parametric
versus nonparametric tests.
7.4 Selecting the Statistical Approach - Parametric Versus Nonparametric Methods
Statistical methods are based on a set of assumptions about the data. The difference between
parametric and nonparametric statistical methods is based on the form and details of these
assumptions. Many statistical tests and models are appropriate only for data that follow a
particular probability distribution. These distributions typically are characterized by one or more
parameters, like the mean or standard deviation, and are called parametric statistical methods.
Parametric methods use the actual data values and assume that the data values follow a specific
probability distribution. Two of the most important distributions for analysis of environmental
data are the normal distribution and the lognormal distribution. If two samples are to be
compared, then parametric methods often require that both sets of data follow the same type of
distribution.
Parametric statistical tests have several distinct advantages over nonparametric tests when the
data follow the assumed distribution. They are often easier to apply and communicate, as they
are based on well-known statistics, such as the mean and standard deviation. On the other hand,
many nonparametric methods rely on the ranks of the sample values or computations involving
all possible pairs of sample values. These calculations become more difficult to implement when
there are a large number of samples to consider. If the sample size is 30 or more, parametric tests
often lead to the same results as their corresponding nonparametric tests unless large outliers are
present.
Parametric tests will have more power than a nonparametric counterpart if the assumptions for
the test are met. However, the distributional assumptions are often strict, and deviation from
these assumptions can lead to misleading results. Parametric tests also have difficulty dealing
with outliers and nondetects. If either is found in the data, then a nonparametric statistical
method may be the preferred approach. In general, nonparametric methods handle outliers and
nondetects better than parametric methods. Nonparametric tests typically use the ranks of the
data, and do not assume that the data follow a specific probability distribution. Because of
reliance on fewer or weaker assumptions about the distribution of the data, nonparametric
methods are often more generally applicable.
This document describes both parametric methods and nonparametric methods. The choice of the
statistical method depends on how the data are distributed. If the distributions of the data appear
to be normal (bell-shaped), then parametric methods are appropriate and have the advantage of
greater power with minimal computations. If the distributions have long tails to the right, than
taking the logarithms of the data is recommended. If the logged values are normally distributed,
Draft Technical Report 172 Revised Draft - November 26, 2012
-------
then parametric methods may be applied to the logged data. If both data sets cannot be
determined to have normal (or lognormal) distributions, then nonparametric methods might be
appropriate.
7.4.1 Determining If Data Have a Normal Distribution
Many statistical tests and models are appropriate only for data that follow a particular
distribution. Methods in this section apply to the normal distribution. The tests for normality also
may be used with lognormal data by applying the prescribed test to the logarithms of the data. If
the data appear to have a distribution other than the normal distribution or the lognormal
distribution, a nonparametric test such as the Wilcoxon Rank Sum test (see Section 7.9.2) should
be considered.
7.4.2 The Shapiro-Wilk W Test
The Shapiro-Wilk W test is considered one of the most powerful tests for normality. This test is
similar to computing a correlation between the quantiles of the standard normal distribution and
the ordered values of a data set. Several EPA guidance documents and many statistical texts
recommend the W test. Tables of critical values for sample sizes up to 50 have been developed
for determining the significance of the test statistic. However, many software packages can
perform the W test for data sets with sample sizes as large as 5,000. This test is difficult to
compute by hand as it requires many summations and multiplications. Therefore, this document
does not give directions for implementing the W test. The Studentized Range test described
below is another test for normality that is much easier to compute, but is not as powerful as the
W test.
7.4.3 The Studentized Range Test
The Studentized Range test for normality is based on the fact that nearly 100% of the area of a
normal curve lies within ±5 standard deviations from the mean. This test uses a ratio of the
sample range to the sample standard deviation. Very large and very small values of the ratio
imply that the data are not well modeled by a normal distribution
This test for normality compares the sample range (maximum value minus minimum value) to
the sample standard deviation. The test statistic is the ratio of the sample range to the standard
deviation. Table 7-17 shows the critical values for determining whether the absolute value of the
ratio for a sample of size TV is significantly too small or too large. If the calculated ratio is less
than the lower tabulated value (a) or exceeds the upper value (b), then the data are not normally
distributed at the specified level of confidence. The Studentized Range test does not perform
well if the data are asymmetric or if the tails of the data are heavier than the normal distribution.
In addition, this test may be sensitive to a few extreme values. The test for outliers described in
the following section may be used to screen the data for outliers before the Studentized Range
test is conducted.
Many environmental data sets are positively skewed with a long but narrow tail of high values
and are similar to a lognormal distribution. If the data appear to be lognormally distributed, then
the range test may be applied to the logged data values.
Draft Technical Report 173 Revised Draft - November 26, 2012
-------
Table 7-17. Critical Values for the Studentized Range Test
N
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
25
30
35
40
45
50
55
60
65
70
75
80
85
90
95
100
150
200
500
1000
Confidence Level
90%
a b
1.78 2.00
2.04 2.41
2.22 2.71
2.37 2.95
2.49 3.14
2.59 3.31
2.68 3.45
2.76 3.57
2.84 3.68
2.90 3.78
2.96 3.87
3.02 3.95
3.07 4.02
3.12 4.09
3.17 4.15
3.21 4.21
3.25 4.27
3.29 4.32
3.45 4.53
3.59 4.70
3.70 4.84
3.79 4.96
3.88 5.06
3.95 5.14
4.02 5.22
4.08 5.29
4.14 5.35
4.19 5.41
4.24 5.46
4.28 5.51
4.33 5.56
4.36 5.60
4.40 5.64
4.44 5.68
4.72 5.96
4.90 6.15
5.49 6.72
5.92 7.11
95%
a b
1.76 2.00
1.98 2.43
2.15 2.75
2.28 3.01
2.40 3.22
2.50 3.40
2.59 3.55
2.67 3.69
2.74 3.80
2.80 3.91
2.86 4.00
2.92 4.09
2.97 4.17
3.01 4.24
3.06 4.31
3.10 4.37
3.14 4.43
3.18 4.49
3.34 4.71
3.47 4.89
3.58 5.04
3.67 5.16
3.75 5.26
3.83 5.35
3.90 5.43
3.96 5.51
4.01 5.57
4.06 5.63
4.11 5.68
4.16 5.73
4.20 5.78
4.24 5.82
4.27 5.86
4.31 5.90
4.59 6.18
4.78 6.39
5.47 6.94
5.79 7.33
99%
a b
1.74 2.00
1.87 2.45
2.02 2.80
2.15 3.10
2.26 3.34
2.35 3.54
2.44 3.72
2.51 3.88
2.58 4.01
2.64 4.13
2.70 4.24
2.75 4.34
2.80 4.44
2.84 4.52
2.88 4.60
2.92 4.67
2.96 4.74
2.99 4.80
3.15 5.06
3.27 5.26
3.38 5.42
3.47 5.56
3.55 5.67
3.62 5.77
3.69 5.86
3.75 5.94
3.80 6.01
3.85 6.07
3.90 6.13
3.94 6.18
3.99 6.23
4.02 6.27
4.06 6.32
4.10 6.36
4.38 6.64
4.59 6.84
5.13 7.42
5.57 7.80
7.5 Outlier Detection
Potential outliers are measurements that are extremely large or small relative to the rest of the
data and may not be representative of the population from which they were collected. Outliers
may result from transcription errors, data-coding errors, or measurement system problems such
Draft Technical Report
174
Revised Draft - November 26, 2012
-------
as errors in chemical analyses. However, it is also possible that outliers may represent true
extreme values of a distribution (hot spots, for instance) and indicate a high degree of variability
in the population. Failure to remove true outliers, or the removal of false outliers, both lead to
distorted estimates of the population parameters. As noted previously, it is essential to remove
true outliers before performing the Studentized Range test for normality. This section discusses
methods to test outliers.
Outlier tests provide statistical evidence that an extreme value does not "fit" the distribution of
the other data and is therefore a potential outlier. Outlier tests should be used only to identify
data points that require further investigation. An outlier test alone cannot determine whether a
statistical outlier should be discarded from a data set. This decision also should be based on
judgmental or scientific considerations such as replicate sampling.
Potential outliers may be identified using graphical methods. Graphs such as histograms, box and
whisker plots, and normal probability plots can all be used to identify observations that are much
larger or smaller than the rest of the data. If potential outliers are identified, the next step is to
apply one of the statistical tests for outliers.
If an outlier is identified, the next step depends on the reason for the outlier. Data points
containing transcription errors should be corrected, while data points collected while an
instrument was malfunctioning may be discarded. Discarding an outlier from a data set should be
done with extreme caution, as environmental data sets often contain legitimate extreme values. If
any data points are found to be outliers through the use of a statistical test, this information
should be documented when the results of the analysis of the data are reported. This information
is critical for subsequent review of the analyses.
7.5.1 Parametric Tolerance Limits for Outliers
A tolerance interval is a range of concentrations containing a pre-specified proportion of the
population of all possible sample values. As the interval is constructed from random samples, a
tolerance interval is expected to contain the specified proportion with only a certain level of
statistical confidence. Two coefficients are required to specify a tolerance interval:
(1) The population proportion that the interval is expected to contain, called the coverage (g).
(2) The level of confidence with which the interval reaches the specified coverage.
A tolerance interval with 95% coverage and a confidence level of 90% is expected to contain, on
average, 95% of the distribution of all possible samples with a probability of 90%.
A "tolerance limit" is a one-sided tolerance interval. Tolerance limits are a parametric statistical
method used for outlier detection. The one-sided upper tolerance limit (UTL) is of most interest
in ground water monitoring. A UTL is designed to be exceeded only in a small percentage of the
measurements. The UTL gauges whether a sample is too extreme relative to the other sample
values and thus should be identified as an outlier.
Draft Technical Report 175 Revised Draft - November 26, 2012
-------
Tolerance limits may be used to identify outliers in the baseline and stabilization periods. For
example, to test the null hypothesis (H0) that a baseline sample is not an outlier, a UTL is
constructed using the baseline data from all wells. Each sample value is then compared to the
UTL to determine if it is an outlier. Specifying a UTL with 95% (or 99%) coverage and a
confidence level of at least 95% is recommended for baseline data sets. A UTL with coverage of
95% (or 99%) would be exceeded, on average by less than 5% (or 1%) of the samples. A
confidence level of 95% indicates that the overall false positive rate for the test is set to
approximately 5%.
Parametric tolerance limits assume normality of the baseline data used to construct the limit. If
the background sample is non-normal, a logarithmic transformation should be applied before
using the tolerance limit. If the log-transformed values appear to be normal, the UTL should be
constructed using the logarithms of the sample values; then, each sample value is compared with
T TTT
e to determine if it is an outlier. Measurements below the limit of detection may be set equal
to the detection limit, or alternatively, may be omitted from the calculations, and the sample size
is adjusted accordingly.
Mathematically, a tolerance limit can be computed with as few as three samples. However, a
sample size of at least eight measurements will be needed to generate an adequate tolerance limit
for outlier detection. Pooling baseline data from multiple wells is recommended to increase the
sample size.
7.5.2 Calculating an Upper Tolerance Limit
Step 1. Calculate the mean M and the standard deviation S of the TV baseline
samples.
Step 2. Construct the one-sided UTL as
UTL=M+kN(g,\-a)-S
where fa(g, 1 - a) is the one-sided normal tolerance factor found in Table 7-18
with a sample size of TV, coverage coefficient of g, and a confidence level of (1 -
a). For other values of TV, use the next lower tabulated value, or calculate
approximate values for fa using linear interpolation.
Step 3. Compare each baseline sample with the UTL. If any sample exceeds the
UTL, there is statistically significant evidence that the sample is an outlier.
Draft Technical Report 176 Revised Draft - November 26, 2012
-------
Table 7-18.
One-Sided Upper Tolerance Limit Factors with g% Coverage
for Selected Values of N
N
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
35
40
45
50
55
60
65
70
75
80
85
90
95
100
95%
g=90%
4.162
3.407
3.006
2.755
2.582
2.454
2.355
2.275
2.210
2.155
2.109
2.068
2.033
2.002
.974
.949
.926
.905
.886
.869
.853
.838
.824
.811
.799
.788
.777
.732
.697
.669
.646
.626
.609
.594
.581
.570
.559
.550
.542
.534
.527
Confidence
95%
5.144
4.203
3.708
3.399
3.187
3.031
2.911
2.815
2.736
2.671
2.614
2.566
2.524
2.486
2.453
2.423
2.396
2.371
2.349
2.328
2.309
2.292
2.275
2.260
2.246
2.232
2.220
2.167
2.125
2.092
2.065
2.042
2.022
2.005
.990
.976
.964
.954
.944
.935
.927
99%
7.042
5.741
5.062
4.642
4.354
4.143
3.981
3.852
3.747
3.659
3.585
3.520
3.464
3.414
3.370
3.331
3.295
3.263
3.233
3.206
3.181
3.158
3.136
3.116
3.098
3.080
3.064
2.995
2.941
2.898
2.862
2.833
2.807
2.785
2.765
2.748
2.733
2.719
2.706
2.695
2.684
99%
g=90%
7.380
5.362
4.411
3.859
3.497
3.240
3.048
2.898
2.777
2.677
2.593
2.521
2.459
2.405
2.357
2.314
2.276
2.241
2.209
2.180
2.154
2.129
2.106
2.085
2.065
2.047
2.030
.957
.902
.857
.821
.790
.764
.741
.722
.704
.688
.674
.661
.650
.639
Confidence
95% <
9.083
6.578
5.406
4.728
4.285
3.972
3.738
3.556
3.410
3.290
3.189
3.102
3.028
2.963
2.905
2.854
2.808
2.766
2.729
2.694
2.662
2.633
2.606
2.581
2.558
2.536
2.515
2.430
2.364
2.312
2.269
2.233
2.202
2.176
2.153
2.132
2.114
2.097
2.082
2.069
2.056
)9%
12.387
8.939
7.335
6.412
5.812
5.389
5.074
4.829
4.633
4.472
4.337
4.222
4.123
4.037
3.960
3.892
3.832
3.777
3.727
3.681
3.640
3.601
3.566
3.533
3.502
3.473
3.447
3.334
3.249
3.180
3.125
3.078
3.038
3.004
2.974
2.947
2.924
2.902
2.883
2.866
2.850
Source: Odeh and Owen (1980)
Draft Technical Report
177
Revised Draft - November 26, 2012
-------
7.6 Determining the Number of Samples per Well
The procedure for calculating the number of post-restoration samples for determining
compliance must address three kinds of comparisons with baseline conditions:
(1) Comparison of post-restoration samples from an individual monitoring well with the
baseline samples from that well
(2) Comparison of post-restoration samples from an individual monitoring well with the
pooled samples from all baseline wells
(3) Comparison of pooled post-restoration samples from all monitoring wells with the pooled
samples all baseline wells
The number of samples may be determined using several approaches. The first, and most basic,
approach is based on the assumptions that the number of samples collected post-restoration is the
same as the number of samples collected in the baseline period, and that the standard deviations
are the same. Let m and GI denote the baseline sample size and standard deviation, and n and 02
denote the post-restoration sample size and standard deviation. The first approach is based on the
assumptions that the sample size is the same in both periods (n=m) and that the standard
deviations are the same (02=01). These constraints preclude most practical applications in
comparisons with different sample sizes in the baseline and post-restoration periods. For each of
the three kinds of comparisons, the sample size and standard deviation will be the same only in
unusual cases.
Generalizing to an approach which permits different sample sizes offers greater flexibility than is
available in the first approach, but at a cost. The cost is additional complexity in the form of a
larger number of tables for determining sample size. This is necessary in order to span the
possible ranges of baseline versus post-restoration sample sizes, as well as several combinations
of the DQO parameters a and P, and a range of possible values for the resolution of the test
MDD/ci.
Further generalization to a third approach which permits both different sample sizes and different
standard deviations offers additional flexibility than is available in the second approach, but with
additional complexity. This is necessary in order to span the possible ranges of baseline versus
post-operational sample variability, in addition to the parameters noted for the second approach.
Power and sample size calculations tend to be much more difficult for nonparametric procedures
than for parametric procedures. Nonparametric procedures usually have less statistical power
than parametric tests when the data follow a known distribution. An adjustment factor of 1.16 is
usually applied to account for the possible loss of efficiency when the nonparametric WRS test is
Draft Technical Report 178 Revised Draft - November 26, 2012
-------
used.21 In addition, MARRSIM (EPA 2000a, Section 5.5.2.4) recommends increasing the
number of samples per well by 20% to account for possible underestimation of o and to prepare
for unplanned events that result in missing or unusable data.
7.6.1 Same Sample Sizes (n=m) and Same Standard Deviation (01=02)
Let «w denote the number of samples to be collected from the post-restoration monitoring wells.
In the first two kinds of comparisons, «w denotes the number of samples to be collected from an
individual post-restoration monitoring well. In the third type of comparison, «w denotes the
number of pooled samples to be collected from all post-restoration monitoring wells.
The number of samples from the post-restoration wells will depend on the selected DQO values
of a and P, and the number of baseline samples (m) that were collected. For the first kind of
comparison, m denotes the number of baseline samples collected from the individual monitoring
well in question. For the second and third kinds of comparisons, m denotes the pooled number of
samples collected from all baseline wells. In this section, it is assumed that the sample sizes are
equal (m=n).
The theory of hypothesis testing provides methods to control the frequency of decision errors
when determining compliance. The decision error rate is reduced by increasing the number of
samples during the post-operational phases. When the sample sizes are equal, the minimum
number of samples per well (nw) to achieve a confidence level of 1 - a and power of 1 - P using
the Mest is obtained using the approximate formula (Campbell et al. 1995):
nw=0.25zla+2(zl_a + z^J a
Here, zp is the pth percentile of the standard normal distribution.
The number of measurements required to achieve the desired decision error rates has a strong
inverse relationship with MDD/o. Smaller values of a and P (leading to larger values for the z
terms) magnify the strength of this inverse relationship. Hence, a tradeoff exists between cost
(number of samples required) and benefit (better power of resolution of the test). This document
does not recommend a specific sample size, as each site will have different variability (o) and
DQO parameters (a and P). A complete set of sample size estimates for nw for
a = 0.01/0.025/0.05/0.10/0.20 and for P = 0.01/0.025/0.05/0.10/0.20 is tabulated for a range of
the MDD/o ratios in Table E-4 in Attachment E. Table E-4 is based on an assumption that the
sample size for the baseline is the same as the sample size for the post-restoration period (m=n)
and that the standard deviations are the same (02=01).
21 The /-test is an optimal test for a shift in the distribution when under the assumption of normality. When
the WRS test is compared to the /-test for a shift in an arbitrary distribution F, the loss in efficiency using the WRS
test is no worse than 108/125 = 0.864,^/or all distributions F (Hodges and Lehman 1956, Theorem 1). The factor of
1.16 (the reciprocal of 0.864) is applied to account for this possible loss of efficiency. On the other hand, the WRS
test can perform much better than the t-test for certain distributions F. A single extreme value may distort the
parametric estimates of the mean and standard deviation required to perform the /-test, while the WRS test may be
unaffected.
22 The value of zp may be calculated in Excel® using the spreadsheet function zp = NORMSINV(p).
Draft Technical Report 179 Revised Draft - November 26, 2012
-------
7.6.2 Different Sample Sizes (n^m), Same Standard Deviations (01=02)
The formula for the number of samples when the sample sizes are unequal is provided by
Campbell et al. (1995). This formula uses the number of samples when the sample sizes are
equal (ww = m = n) given in Section 7.6.1 as the starting point for determining unequal sample
sizes. When m is known in advance and is larger than nw/2, the number of samples per well in the
post-restoration period is given by:
When the number of baseline samples m is smaller than «w/2, there is no value of n that will
achieve the DQO parameters a, P, andMDD/c\ used to calculate the value of ww.
When the number of samples in the baseline data set is larger than ww, the number of samples
required in the post-restoration period will be smaller than ww, and vice versa. The total number
of samples (n+rri) will be greater than 2«w when the sample sizes are unequal. This occurs
because the use of equal sample sizes is the optimal allocation for achieving the DQO parameters
with a minimum number of samples.
The values of nw are provided for 9 alternative values for a and P in Tables E-5 through E-13.
The combinations of the DQO parameters a and P are provided for (a, P) =
(0.05, 0.05), (0.05, 0.10), (0.05, 0.20);
(0.10, 0.05), (0.10, 0.10), (0.10, 0.20),
(0.20, 0.05), (0.20, 0.10), (0.20, 0.20);
The rows of these tables show the number of samples that were collected in the baseline (m). For
the indicated value of m, the columns of the table are indexed by the desired level of resolution
MDD/o.
7.6.3 Different Sample Sizes (n^m) and Different Standard Deviations (01^02)
Generalizing to a model which permits different sample sizes offers greater flexibility than is
available in the approach underlying Table E-4, but at a cost. The cost is additional complexity in
the form of a larger number of tables for determining sample size. This is necessary to span the
wide possible ranges of baseline versus post-restoration sample sizes, baseline versus post-
operational sample variability, as well as several combinations of the DQO parameters a and P
and the test resolution
If there are m baseline samples with standard deviation GI, the desired degree of resolution for
the test isMDD/oi, and the post-restoration samples are estimated to have standard deviation 02,
then the number of samples «w is the smallest value of n satisfying the inequality:
1 V2 \ MDD
— H <
m n
Draft Technical Report 180 Revised Draft - November 26, 2012
-------
where v = a 2 /a l is the ratio of the standard deviations. Here z\.a and zi_p have the standard
definitions.
When the number of baseline samples is known, the equation may be solved for the number of
post-restoration samples required to achieve the DQO parameters. An iterative solution of the
equation is found by using a series of trial values of n starting at 1 and continuing until the
inequality is achieved. The expression on the left of the inequality decreases asymptotically to
(zl_a + Zj_p )/\m as n increases. Hence, if MDD/al < (zl_a + zl_p)/y/m , there is no n sufficiently
large to achieve the desired test resolution ofMDD/ci given the specified values of m, a, and p.
Table E-14 shows the number of post-restoration samples «w for the DQO parameters (a,P) =
(0.05,0.05). The rows in this table are segmented by the number of baseline samples m. For the
selected value of m, the table includes rows for all achievable values ofMDD/c\. The columns of
the table indicate the ratio of the post-operational to baseline standard deviation: v = a 2 /a l .
Large sample sizes may be encountered in pooled comparisons in large fields. The table shows
post-restoration sample size estimates up to 1,000. Estimates exceeding this level are indicated
by an asterisk.
Only Table E-14 is included in the current Attachment E. Eight other tables showing the number
of samples «w for other combinations of the DQO parameters (a, P) =
(0.05, 0.10), (0.05, 0.20);
(0.10, 0.05), (0.10, 0.10), (0.10, 0.20),
(0.20, 0.05), (0.20, 0.10), (0.20, 0.20),
have been prepared. As these tables are voluminous, they have not been included in
Attachment E, but can be provided on request.
7.7 Statistical Methods for Trends and Seasonally
The existence of seasonality may complicate trend analysis and comparisons with baseline
conditions. Sufficient data must be collected to estimate seasonal trends and account for
seasonality in the statistical analysis. This requires at least two years of data (a minimum of one
year during the pre-operational phase plus one year during the post-operational phase) under the
assumption the seasonal pattern is not affected by the mining and restoration activities.
Sampling locations should be monitored for at least a full year prior to operations, as well as
after restoration, with consistent timing of observations within these years. A carefully designed
monitoring plan, in which each well has equivalently timed baseline and post-restoration
measurements (quarterly or monthly measurements taken at the same periods in each quarter or
month), will largely eliminate the need for seasonal adjustment when statistical tests of post/pre-
operational differences are performed. Use of consistent timing in both periods has the effect of
"subtracting out" any seasonal component when post-restoration data are compared to the initial
conditions.
Draft Technical Report 181 Revised Draft - November 26, 2012
-------
If seasonal patterns are either predicted by hydrological models or observed to be highly variable
from year to year, then a single year of pre-operational information may not be adequate. A
design in which pre-operational data are collected for more than 1 year may be preferable for
ISR operations within shallow aquifers. This is especially true in unusual years of extreme
drought or flooding, for example, that may bias the baseline measurements one way or another.
One approach for extending the time seasonality is monitored is the use of additional monitoring
data collected outside the ore zone. Up gradient wells may be monitored for seasonal effects
throughout operational period in lieu of additional pre-operational monitoring.
The first step in analyzing measurements in one or more wells is to plot the data as a time series.
Such plots are shown in the example discussed in Attachment D. Plots of the data may reveal
patterns such as seasonality and/or the existence of outliers or blunders in the data. Outliers are
values that appear to be unusually high or low when compared to the other values. Outliers may
be valid data or may arise from unusual circumstances unrelated to the process being measured.
Blunders are outright errors made in recording the data, transcription, or calculations. A common
blunder is a mistake in the units of measure. Plotting is used to detect these situations, but does
not provide for an explanation or resolution for the unusual value. Blunders are sometimes
outliers and thus a given value could be both. But blunders need not be outliers—data may have
been keyed in wrong but not be far from the main body of observations. So plotting will not
always reveal blunders in the data.
If a value is identified as erroneous, it should be removed from the data set. In cases of doubt, the
value should be retained. The nonparametric statistical tests discussed in this section were
selected because of their robustness. The statistical term "robust" is loosely defined as resistant
to the effects of outliers and blunders in the data.
7.7.1 Adjusting for Seasonality
Seasonality may occur in baseline samples in Phase 1, while the site is reaching steady state in
Phase 4 and/or in Phase 5, where seasonality may affect decisions concerning long-term stability
and whether target remediation values are attained.
Seasonality is a pattern that repeats periodically in a cycle. An annual seasonal pattern has a
cycle that spans 12 months or 4 quarters. A seasonal index measures how far the average for a
particular period is above (or below) the average for all periods. The unified guidance for RCRA
(EPA 2009) recommends the following concerning seasonality:
Seasonal fluctuations in intrawell background can be treated in one of two ways.
A seasonal Mann-Kendall trend test built to accommodate such fluctuations can
be employed (Section 14.3.4). Otherwise, the seasonal pattern can be estimated
and removed from the background data, leaving a set of seasonally-adjusted data
to be analyzed with either a prediction limit or control chart. In this latter
approach, the same seasonal pattern needs to be extrapolated beyond the current
background to more recent measurements from the compliance well being tested.
These later observations also need to be seasonally-adjusted prior to comparison
against the adjusted background, even if there is not enough compliance data yet
collected to observe the same seasonal cycles.
Draft Technical Report 182 Revised Draft - November 26, 2012
-------
However, the guidance adds the following caveat:
Corrections for seasonality should be used cautiously, as they represent
extrapolation into the future. There should be a good physical explanation for the
seasonal fluctuation as well as good empirical evidence for seasonality before
corrections are made. Higher than average rainfall for two or three Augusts in a
row does not justify the belief that there will never be a drought in August, and
this idea extends directly to groundwater quality. At least three complete cycles of
the seasonal pattern should be observed on a time series plot before attempting
the adjustment below. If seasonality is suspected but the pattern is complicated,
the user should seek the help of a professional statistician.
The seasonal Mann-Kendall test is a variation of the Mann-Kendall test for trends described
below in Section 7.7.2.1. The test is described in detail in EPA 2009 in Section 14.3.4.
Seasonal adjustment procedures are commonly applied to ecological and economic data to
account for seasonal patterns. The process of "deseasonalizing" the data removes these periodic
seasonal variations to reveal the underlying longer term pattern. The /'th seasonal component (Q;)
is defined as the deviation of the seasonal mean (Y;) from the overall mean (YM): Qi = Y; - YM.
The deseasonalized time series (X) is obtained by subtracting the seasonal means from the
original data series: Xt;; = Yt;; - Q; (EPA 2009, Eq. 14.23). The deseasonalized data series has the
short-term seasonal variations removed; longer-term trends remain in the data. Plots of the
seasonally adjusted data series are useful for determining when suspected outliers in sample
values reflect the normal variability of monitored parameters after adjusting for the seasonal
variations.
When there are four quarterly measurements in each year, the data may be seasonally adjusted by
the procedure described in Section D.I in Attachment D. Appropriate modifications must be
made for periodic variations based on other time frames. Some parameters may require seasonal
adjustment and others not. Formal tests for the presence of seasonality across several wells are
based on an analysis of variance. This procedure is described in EPA 2002b (Sections 14.2.2
and 14.3.3).
The seasonal adjustment procedures are applicable to data that are approximately symmetric and
normally distributed. For highly skewed lognormal data series, the calculations above would be
applied to the logarithms of the measurements. This is equivalent to using the ratio of the
quarterly mean to the overall mean (Q i=Y;/YM) as the seasonality index in place of the additive
index above. If this index is 1.2, this means that, on average, the period (season) is 20% higher
than average. In this case, the seasonally adjusted data series is obtained by dividing the original
data series by the seasonal index: Xy = Y^/Q*;.
It is assumed that there is a complete set of quarterly measurements for 3 years with no missing
or nondetect values. If one or two nondetects occur in the data series, one should replace those
values with the limit of detection. If there is at most one missing data value, these methods may
be applied using the averages of the available data to compute the seasonal index. If more than
one value is missing, the appropriateness of adjusting for seasonal variation should be discussed
with a statistician familiar with environmental sampling.
Draft Technical Report 183 Revised Draft - November 26, 2012
-------
Unless otherwise noted, in the remaining sections of this chapter, the term
"data" refers to the seasonally adjusted data series Xtti.
7.7.2 Using Trend Tests to Determine Stability
Mining by ISR causes a major perturbation of the physical and chemical environment of the ore
zone. When mining operations end and restoration activities are completed, the ground water and
the minerals it contacts begin to shift toward a new geochemical steady state. It should be
recognized that the system may never return to steady-state conditions exactly comparable to
pre-operational baseline conditions.
In the short run, post-restoration samples are used to provide statistical evidence of trends or lack
thereof. However, the long-term stability of the site is best understood using a conceptual model
that considers these data along with the hydrology, contaminant transport, and geochemical
reactions to provide a qualitative estimate of the evolution of the system after post-restoration
monitoring is completed. The NRC (2007) has emphasized that development of a justifiable
conceptual model capturing the major chemical and physical phenomena at each site is required.
This approach allows for site-specific flexibility.
In addition to water quality data, mineralogical data can be used to evaluate the long-term
stability of the system. Hydro-geochemical modeling can be a valuable tool for modeling the rate
of the return to stability and for predicting how long a system will take to return to baseline
conditions. Although complete mineralogical characterization may require additional resources,
a fully developed quantitative model supported by site post-restoration measurements can
provide additional confidence that the restoration goal of site stability after closure has been met.
Statistical tests for trends are used to demonstrate stability within the specified monitoring
period. These tests may be used with any time series of four or more independent samples to test
for trends in well parameters. Trend tests are employed in Phase 1 to check for unexpected trends
in baseline samples, and particularly in Phase 5 to affirm long-term stability. Trends may be
detected using parametric and/or nonparametric statistical tests. The power of a trend test to
detect a trend depends on several factors, including the underlying variability of the series, the
magnitude of slope we wish to detect, level of confidence for the test and the length of time the
series is observed.
A Monte Carlo simulation was conducted to examine the time it takes to detect a trend. Two
9^
popular methods of trend detection are considered: the linear regression t-test and the Mann-
Kendall test. The regression r-test is a parametric statistical test, and the Mann-Kendall test is a
nonparametric test.
23 A complete discussion of linear regression techniques for assessing trends and projecting probable future
levels is found in EPA 1992, Chapter 6.
Draft Technical Report 184 Revised Draft - November 26, 2012
-------
7. 7.2.7 Detecting Trends Using Hypothesis Tests
Given a series of T sample values xl , • • • , XT collected from a well at times tl , • • • , tT , the obj ective is
to decide if there is a significant trend in the series. In general, time may be measured in any
units, such as months, quarters, or years. However, the interpretation of the slope of the trend
line will depend on the units selected for the time axis as well as the units of measurement for the
sample values. In this discussion, quarterly sampling is assumed as quarterly measurements are
commonly used. Sampling frequencies other than quarterly sampling are discussed in
Section 7.7.2.4. The time variable is measured in number of years from the first observation. The
slope estimates are expressed as a percentage change per year. The simulations assume that the
sample collection times are equally spaced, one per quarter. In general, this is not a necessary
assumption. Both methods of trend detection may be applied with irregularly spaced sampling
times.
The linear trend model provides a parametric statistical test for a significant trend in the
percentage deviations. A linear trend is described by a trend line with the equation:
Here the coefficient a is the initial value of y at time t=0, often described as the intercept of the
trend line. The coefficient b is the slope of the trend line. If time is increased by one unit, the
value of y is increased (or decreased) by b units when b is positive (or negative).
The regression t-tesi is used to test if an observed positive or negative slope in the series is
statistically significant at a specified level of confidence. The ^-statistic for the test is the slope
/\ /\ / /\
estimate i> divided by its standard error of estimation: t(b) = bj SE(b). If the absolute value of the
^-statistic for the slope coefficient is larger than the critical value for the test, then the trend has a
significantly positive (or negative) slope. If the slope of the trend line is not zero, all statistical
tests will detect the trend eventually. In the short term, the test may or may not detect a trend.
The power to detect a trend will depend on the magnitude of the trend, the degree of temporal
variability, the level of confidence for the test, and how long we are willing to wait to detect the
trend. A complete discussion of the use of linear and nonlinear regression models for assessing
trends is provided in Chapter 6 of EPA 1992.
7. 7. 2. 2 Simulation of Trend Detection
A Monte Carlo simulation study was conducted to estimate the time required to detect a linear
trend using regression ^-test and the Mann-Kendall trend test. In the simulation study, a time
series of T quarterly samples is generated for each combination of slope and variability.
Quarterly samples were generated at times /j,y'=l,. . ., T where the time t is measured in years
from the first sample. For example, with five quarters of data (r=5), the five values of tj are t\=0,
^2=0.25, ^3=0.5, ^4=0.75, and ^=1.0 years. This section presents a summary of the simulation
results. Details of the simulation analysis are provided in Attachment G.
Draft Technical Report 185 Revised Draft - November 26, 2012
-------
Regression analysis provides an estimate of the magnitude of the trend, and the regression M
is used to determine if the trend is statistically significant. The Mann-Kendall test is a
nonparametric test for detecting trends in the data series. The Mann-Kendall test may be used
with any series of four or more independent samples to test for trends. The Mann-Kendall test
does not provide an estimate of the slope, however; only a test of whether the trend is significant.
The Mann-Kendall test and the regression t-test have very similar characteristics when the
observation errors are normally distributed. Regression analysis is sensitive to outliers and
requires careful inspection of the data before the results of the Mest can be validated. Regression
also requires numerical values for all samples, and pseudovalues must be assigned for
nondetects. The Mann-Kendall test is not unduly influenced by outliers and is therefore more
robust than regression for detecting trends. The Mann-Kendall test also may be used with
nondetects.
Table 7-19 through Table 7-21 show the results of the simulations. The tables contain results for
11 selected values of the slope that is to be detected, ranging from 1% per year to 100% per year.
For each slope, 9 levels of variability are considered ranging from 5% to 150%. Each table
shows the minimum number of quarterly samples required, such that the probability of detecting
the slope is at least 90%, 95% or 99%, respectively.
In the upper right corner of the Mann-Kendall tables, an asterisk is used when the number of
samples required exceeds 100. In these regions of the tables, the variability is too large to detect
the specified slope with less than 25 years of quarterly samples. The regression tables do not
have this constraint, and the series were permitted to be as long as necessary to detect the slope.
Using Table 7-20 for a 95% chance of detection using regression, 21 samples are required when
the slope is 10% and the variability is 20%. The same number of samples (21) is required for a
slope of 15% and a variability of 30%. The corresponding number of samples using the Mann-
Kendall test is 22 for both cases. These two cases are similar in that the ratio of the slope to the
variability is 0.5.
The simulation results suggest that linear trends will be detected using the Mann-Kendall test
within 15 quarters in 95% or more of the cases if the standardized slope ratio is greater than or
equal to 1. When using the regression t-test under the most favorable assumptions, linear trends
may be detected within 14 quarters in 95% or more of the cases if the standardized slope ratio is
greater than or equal to 1. Doubling of the number of samples to 30 (or 28 for regression)
enables detection of slopes equal approximately to 1/3 of the variability.
Draft Technical Report 186 Revised Draft - November 26, 2012
-------
Table 7-19. Number of Quarterly Samples Required for 90% Probability of Detecting Slope Using a Mann-Kendall
or Regression Trend Test
Regression Trend Test
Slope
(%/year)
1
3
5
10
15
20
30
40
50
75
100
Variability (%)
5 10 15 20 30 50 75 100 150
36 56 73 88 114 161 210 254 333
18 27 35 43 56 78 101 122 159
13 20 26 31 40 56 72 87 114
9 13 17 20 26 36 46 56 73
7 10 13 15 20 27 35 43 56
6 9 11 13 16 23 30 35 46
5 7 9 10 13 18 23 27 35
5 6 7 9 11 15 19 23 30
4 6 7 8 10 13 16 20 26
45568 10 13 15 20
44557 9 11 13 17
Mann-Kendall Trend Test
Slope (%/year)
1
3
5
10
15
20
30
40
50
75
100
Variability (%)
5 10 15 20 30 50 75 100 150
37 57 74 90 * * * * *
18 28 37 44 57 78 * * *
13 20 26 32 41 57 75 89 *
9 13 17 20 27 36 47 57 74
8 11 13 16 21 28 36 44 57
7 9 12 13 17 24 30 37 47
6 8 9 11 14 18 24 28 37
5 7 8 9 11 15 20 24 31
5 6 7 9 10 13 17 20 27
4 5 6 7 9 11 13 16 20
4 5 5 6 7 9 12 13 17
* Over 100 quarters. Mann-Kendall test simulation is limited to 25 years.
Draft Technical Report
187
Revised Draft - November 26, 2012
-------
Table 7-20. Number of Quarterly Samples Required for 95% Probability of Detecting Slope Using a Mann-Kendall
or Regression Trend Test
Regression Trend Test
Slope
(%/year)
1
3
5
10
15
20
30
40
50
75
100
Variability (%)
5 10 15 20 30 50 75 100 150
38 61 79 95 124 175 225 271 358
19 30 38 46 60 84 109 133 175
14 21 27 33 44 60 79 96 123
9 14 18 21 28 39 50 60 79
7 11 14 17 21 30 38 46 60
6 9 12 14 18 25 32 39 50
5 7 9 11 14 19 24 30 38
5689 11 16 20 25 32
5 6 7 8 10 14 18 21 27
4567 8 11 14 17 21
4456 7 9 12 14 18
Mann-Kendall Trend Test
Slope (%/year)
1
3
5
10
15
20
30
40
50
75
100
Variability (%)
5 10 15 20 30 50 75 100 150
39 62 80 97 * * * * *
20 30 39 47 62 85 * * *
14 22 29 34 44 62 80 97 *
10 14 18 22 29 40 51 62 80
8 12 14 17 22 31 39 47 61
7 10 12 14 19 25 32 39 52
6 8 10 12 15 20 25 31 39
5 7 9 10 12 16 21 25 33
5 7 8 9 11 14 18 22 28
4 5 7 7 9 12 14 17 22
4 5 6 7 8 10 12 14 18
* Over 100 quarters. Mann-Kendall test simulation is limited to 25 years.
Draft Technical Report
188
Revised Draft - November 26, 2012
-------
Table 7-21. Number of Quarterly Samples Required for 99% Probability of Detecting Slope Using a Mann-Kendall
or Regression Trend Test
Regression Trend Test
Slope
(%/year)
1
3
5
10
15
20
30
40
50
75
100
Variability (%)
5 10 15 20 30 50 75 100 150
43 68 89 106 139 197 258 306 405
21 34 44 52 68 95 123 150 198
16 24 31 37 49 68 88 108 136
10 15 20 24 32 43 56 69 89
8 12 15 19 24 33 43 52 68
7 10 13 15 20 28 36 43 56
6 8 10 12 15 21 28 33 43
5 7 9 10 13 18 23 28 36
568 9 11 16 20 24 31
456 7 9 12 15 19 24
456 6 8 10 13 16 20
Mann-Kendall Trend Test
Slope (%/year)
1
3
5
10
15
20
30
40
50
75
100
Variability (%)
5 10 15 20 30 50 75 100 150
45 70 91**** * *
22 34 44 54 70 96 * * *
16 25 32 39 50 70 90 * *
11 16 21 25 32 45 58 69 89
9 13 16 20 25 34 45 53 69
8 11 14 16 21 29 37 44 58
7 9 11 13 16 22 29 34 45
6 8 10 11 14 18 24 28 38
5 7 9 10 12 16 21 25 32
5 6 7 8 10 13 16 19 25
4 5 7 7 9 11 14 16 20
* Over 100 quarters. Mann-Kendall test simulation is limited to 25 years.
Draft Technical Report
189
Revised Draft - November 26, 2012
-------
7.7.2.3 Duration of Sampling
Table 7-20 shows the required number of samples to achieve a 95% probability of detection
using the regression Mest or the Mann-Kendall test for selected slope and variability. The
simulation results also provide estimates of the probability of detecting a trend using a fixed
number of quarterly samples. This section presents a comparison of three alternatives: quarterly
sampling for 12 quarters (3 years); 20 quarters (5 years); or 32 quarters (8 years). The
comparison uses the same set of slopes and variability as in Table 7-20.
Table 7-22 shows the probability of detecting a trend (expressed as a percentage) using the
Mann-Kendall trend test or the regression t-tesi with 12 quarterly samples. The probability of
detection was simulated for the combinations of slope and variability shown on the borders of
the tables. The slope is expressed as the percentage change per year, and the variability is
expressed in terms of the standard deviation from the trend line. The outlined cells on the
diagonal of the table indicate where the slope is equal to the variability.
Using 12 quarterly samples, there is approximately an 80% chance of detecting the trend using
the Mann-Kendall test when the slope is equal to the variability. When the regression t-tesi is
used with normally distributed observation errors, the chance of detection increases to
approximately 87% when the slope is equal to the variability. When the slope is less than the
variability, the probability of detection is always smaller than these values, falling quickly to
below 50%. The apparently better performance of the regression t-test over the Mann-Kendall
test is largely due to the fact that the simulation uses normally distributed errors.
Table 7-23 shows simulation results for 20 quarterly samples spread over 5 years. There is
approximately a 100% chance of detecting the trend when the slope is equal to the variability.
Table 7-24 shows simulation results for 32 quarterly samples (8 years). In this case, the chance
of detecting slopes as small as one-half of the variability is approximately 100%.
Draft Technical Report 190 Revised Draft - November 26, 2012
-------
Table 7-22. Probability of Detecting a Trend with 12 Quarterly Samples
Mann-Kendall Trend
Slope
(%/yr)
1
3
5
10
15
20
30
40
50
75
100
Test
Variability (%)
5
12
45
80
100
100
100
100
100
100
100
100
10
7
18
35
80
98
100
100
100
100
100
100
15
6
12
20
50
80
96
100
100
100
100
100
20
6
9
15
33
60
82
98
100
100
100
100
30
6
8
10
21
35
51
81
96
99
100
100
50
5
6
7
12
17
25
45
64
81
98
100
75
5
6
6
9
12
16
25
38
52
81
96
100
5
5
6
7
9
12
17
25
33
61
81
150
4
5
5
7
8
9
12
16
20
34
51
Regression f-test for Trend
Slope
(%/yr)
1
3
5
10
15
20
30
40
50
75
100
Variability (%)
5
14
51
87
100
100
100
100
100
100
100
100
10
9
21
40
87
99
100
100
100
100
100
100
15
7
14
24
58
87
98
100
100
100
100
100
20
7
11
17
39
67
87
99
100
100
100
100
30
6
9
12
24
40
59
87
98
100
100
100
50
6
7
8
15
20
29
51
72
87
100
100
75
6
7
7
11
14
19
29
44
59
87
98
100
5
6
7
8
11
14
21
30
39
68
89
150
5
6
6
8
9
10
14
19
24
40
58
Key:
Less than 50% chance of detecting trend.
Between 50% and 95% chance of detecting trend.
Unshaded Greater than 95% chance of detecting trend.
Outlined Outlined cells have slope equal to variability.
Draft Technical Report
191
Revised Draft - November 26, 2012
-------
Table 7-23. Probability of Detecting a Trend with 20 Quarterly Samples
Mann-Kendall
Slope
(%/yr)
1
3
5
10
15
20
30
40
50
75
100
Trend Test
Variability (%)
5
31
97
100
100
100
100
100
100
100
100
100
10
14
56
91
100
100
100
100
100
100
100
100
15
11
33
63
99
100
100
100
100
100
100
100
20
9
22
43
91
100
100
100
100
100
100
100
30
7
14
26
62
90
99
100
100
100
100
100
50
6
10
15
32
55
76
97
100
100
100
100
75
6
8
11
19
32
47
77
94
99
100
100
100
5
7
9
15
22
32
55
76
91
100
100
150
6
6
6
11
15
19
33
47
63
91
99
Regression f-test for Trend
Slope
(%/yr)
1
3
5
10
15
20
30
40
50
75
100
Variability (%)
5
34
98
100
100
100
100
100
100
100
100
100
10
16
59
93
100
100
100
100
100
100
100
100
15
11
34
66
99
100
100
100
100
100
100
100
20
9
23
46
93
100
100
100
100
100
100
100
30
7
15
27
66
92
99
100
100
100
100
100
50
6
10
15
33
58
80
98
100
100
100
100
75
6
8
12
20
34
50
81
95
99
100
100
100
5
7
10
15
23
35
58
79
93
100
100
150
6
6
7
12
16
20
34
51
67
93
100
Key:
Less than 50% chance of detecting trend.
Between 50% and 95% chance of detecting trend.
Unshaded Greater than 95% chance of detecting trend.
Outlined Outlined cells have slope equal to variability.
Draft Technical Report
192
Revised Draft - November 26, 2012
-------
Table 7-24. Probability of Detecting a Trend with 32 Quarterly Samples
Mann-Kendall
Slope
(%/yr)
1
3
5
10
15
20
30
40
50
75
100
Trend Test
Variability (%)
5
78
100
100
100
100
100
100
100
100
100
100
10
33
98
100
100
100
100
100
100
100
100
100
15
19
78
99
100
100
100
100
100
100
100
100
20
14
58
92
100
100
100
100
100
100
100
100
30
10
33
66
99
100
100
100
100
100
100
100
50
8
19
33
78
98
100
100
100
100
100
100
Key:
Unshaded
Outlined
75
7
12
19
49
78
95
100
100
100
100
100
100
6
10
14
33
58
78
98
100
100
100
100
150
5
8
9
20
33
48
80
95
99
100
100
Regression t-test for Trend
Slope
(%/yr)
1
3
5
10
15
20
30
40
50
75
100
Variability (%)
5
81
100
100
100
100
100
100
100
100
100
100
10
36
99
100
100
100
100
100
100
100
100
100
15
21
82
100
100
100
100
100
100
100
100
100
20
15
61
94
100
100
100
100
100
100
100
100
30
10
35
70
99
100
100
100
100
100
100
100
50
9
20
35
81
99
100
100
100
100
100
100
75
7
13
21
52
81
97
100
100
100
100
100
100
7
10
16
35
61
82
98
100
100
100
100
150
6
8
10
22
35
53
83
96
100
100
100
Less than 50% chance of detecting trend.
Between 50% and 95% chance of detecting trend.
Greater than 95% chance of detecting trend.
Outlined
cells have slope equal to variability.
Draft Technical Report
193
Revised Draft - November 26, 2012
-------
7.7.2.4 Sampling Frequency
When the sampling frequency changes from the base case of one sample per quarter to a higher
or lower frequency (say bi-annual or monthly), there are several important factors to consider.
Higher sampling frequencies increase the possibility that there will be correlations between
successive samples. The simulation model described above does not consider correlations in the
series of samples. If there are possible correlations, the required number of samples is larger than
the simulation predicts. In the unlikely case of negative correlations, the required number of
samples would be smaller than the simulation predicts. In the following discussion, correlation is
ignored.
In the absence of correlation, higher sampling frequencies affect the required number of samples
in two opposite ways. A higher sampling frequency provides a greater number of samples, but in
a shorter period of time. The effect of a greater number of samples is clear: it will increase the
power of detection. However, a shorter period of observation means that the time series does not
have as long to change and the change that has occurred is more difficult to detect.
To explore the implications of higher or lower sampling frequencies in the absence of
correlation, one case in Table 7-20 was selected for a "one-off analysis. Table 7-20 was
generated using 4 quarterly samples per year. The case in Table 7-20 for a 95% chance of
detecting a slope of 10%/year and a variability of 20% was selected for this analysis. This case
requires 21 samples for a 95% chance of detection using regression, and 22 samples using the
Mann-Kendall test when sampling is conducted quarterly.
Sampling frequencies of 2 per year (semiannual), 4 per year (quarterly), 8 per year
(approximately 6-week intervals), and 12 per year (monthly) were selected for the simulation
analysis. Table 7-25 shows results of the simulation. The selected case from Table 7-20 is
highlighted in this table at a sampling frequency of 4 samples per year.
When the sampling frequency is reduced to semiannual (2 per year), the required number of
samples for detecting a trend using the regression r-test is reduced from 21 quarterly samples to
14 semiannual samples. It will take 7 years to collect the 14 semiannual samples, compared with
5.25 years to collect 21 quarterly samples. Results for the regression Mest and the Mann-Kendall
test are almost identical. In the absence of correlation, less frequent sampling requires fewer
sampling events, but the samples are spread over a longer period of time. It is more unlikely to
encounter correlations using semiannual data than using quarterly data, so this result of the one-
off analysis is reasonably robust with respect to the assumed lack of correlation.
If sampling is conducted more frequently than quarterly, more samples are required. If a 6-week
sampling interval is used (approximately 8 per year), the required number of samples for
detecting a trend using regression is increased from 21 quarterly samples to 33 samples. It will
take slightly over 4 years to collect the 33 six-week samples, compared with 5.25 years to collect
21 quarterly samples. In the absence of correlation, more frequent sampling requires more
sampling events, but the samples are spread over a shorter period of time. If a monthly sampling
interval is used (12 per year), the required number of samples for detecting a trend using
regression is increased to 43 samples. It will take approximately 3.5 years to collect the 43
Draft Technical Report 194 Revised Draft - November 26, 2012
-------
monthly samples. It should be noted that monthly sampling is the most likely to encounter
autocorrelation, which would increase the required number of samples and number of years
above the predictions made by the current simulation model.
Figure 7-14 contains a plot comparing the number of samples and number of years required to
have a 95% chance of detection using the regression t-test with the sampling frequencies shown
in Table 7-25. The point where the two lines cross is the case highlighted in the table. This plot
shows the trade-off encountered when the sampling frequency is increased by a factor of 6 from
semiannual to monthly. The required number of samples increases by a factor of approximately
3, while the duration of the sampling is decreased by less than half.
Table 7-25. Number of Samples and Number of Years Required for 95% Chance of
Detection Using Regression or Mann-Kendall Test
(Case: Slope = 10%/year and Variability = 20%)
Number of
Samples
Number of
Years
Sampling
Frequency
2 per year
4 per year
8 per year
12 per year
2 per year
4 per year
8 per year
12 per year
Regression
14
21*
33
43
7
5.25*
4.13
3.58
Mann-Kendall
14
22*
34
44
7
5.5*
4.25
3.67
*Case in Table G-2 for slope of 10%/year and variability of 20%.
Draft Technical Report
195
Revised Draft - November 26, 2012
-------
40
o
8 c
Si
5 8
•^ »-
O O)
II20
15
3 4 5 6 7 8 9 10 11 12 13
Sampling Frequency (#/year)
# of Samples -e— # of Quarters
Figure 7-14. Plot of Number of Samples and Number of Quarters Required for 95%
Chance of Detection Using Regression versus Sampling Frequency
(Case: Slope = 10%/year and Variability = 20%)
7.7.2.5 Caveats Concerning Regression Analysis of Trends
The linear regression model provides estimates of the slope of the trend line and its standard
error based on least squares. The ratio of these estimates is the ^-statistic used in the regression
Mest for a significant slope. Many parametric tests like the linear regression Mest are based on
the normality assumption or equivalent. These tests perform well when the observation errors are
normally distributed. When the observation errors are not normally distributed and include
random outliers, the least squares estimates of the slope of the regression line and its standard
error may be unduly influenced by these outliers.
When regression analysis is used, it is important to verify the assumptions, including:
• Normality of residuals
• Equal variances
• Independence
• Sensitivity to outliers
The Mann-Kendall test is a nonparametric test for trends. Both tests have very similar
characteristics when the observation errors are normally distributed. This fact is confirmed by the
similarity of simulation results for these tests in Table 7-19 through Table 7-24. Unlike the
Draft Technical Report
196
Revised Draft - November 26, 2012
-------
regression Mest, the Mann-Kendall test also is expected to perform well when the errors of
observation are not normally distributed.
The two tests also differ in the treatment of nondetect values. The Mann-Kendall test is based on
counts of how many observations are higher, lower, or the same as the ones before. When the
time series contains nondetect values, the counts are made by treating all nondetects as equal and
assuming all other values in the series are greater than the nondetects. The counts may be made
without assuming a specific value for the nondetects. Such is not the case with parametric
models. The most elementary parametric estimate is to use the arithmetic average as an estimate
of the mean. This simple estimate cannot be calculated without assuming values for the
nondetects. Estimation using linear regression and least squares is only possible if specific values
are used for the nondetects, usually one-half of the detection level.
Despite the advantages of the Mann-Kendall test for detecting trends, the test does not provide an
estimate of the magnitude of the slope. Other nonparametric methods are available for estimating
the slope, including the Theil-Sen trend line estimator.24
The following text was excerpted from EPA QA/G-9S (EPA 2006a):
4.3.2.1 Estimating a Trend Using the Slope of the Regression Line
The classic procedures for assessing linear trends involve regression. Linear
regression is a commonly used procedure in which calculations are performed on
a data set containing pairs of observations (Xt, Yj), so as to obtain the slope and
intercept of a line that best fits the data. For temporal data, the Xt values
represent time and the Yt values represent the observations. An estimate of the
magnitude of trend can be obtained by performing a regression of the data versus
time and using the slope of the regression line as the measure of the strength of
the trend.
Regression procedures are easy to apply. All statistical software packages and
spreadsheet programs will calculate the slope and intercept of the best fitting line,
as well as the correlation coefficient r (see Section 2.2.4). However, regression
entails several limitations and assumptions. First of all, simple linear regression
(the most commonly used method) is designed to detect linear relationships
between two variables; other types of regression models are generally needed to
detect non-linear relationships such as cyclical or non-monotonic trends.
Regression is very sensitive to outliers and presents difficulties in handling data
below the detection limit, which are commonly encountered in environmental
studies. Hypothesis testing for linear regression also relies on two key
assumptions: normally distributed errors, and constant variance. It may be
difficult or burdensome to verify these assumptions in practice, so the accuracy of
the slope estimate may be suspect. Moreover, the analyst must ensure that time
plots of the data show no cyclical pattern; outlier tests show no extreme data
24 The Theil-Sen trend estimator is described in detail in EPA 2009 in Section 17.3.3. The Mann-Kendall
test is also described in EPA 2009 in Section 17.3.2.
Draft Technical Report 197 Revised Draft - November 26, 2012
-------
values; and data validation reports indicate that nearly all the measurements
were above detection limits. Due to these drawbacks, linear regression is not
recommended as a general tool for estimating and detecting trends, although it
may be useful as an informal and quick screening tool for identifying strong
linear trends. [Emphasis added.]
The nonparametric Mann-Kendall test for trends is recommended in this document for detecting
trends. However, the Mann-Kendall test does not provide an estimate of the magnitude of the
trend. Once a trend has been detected using the Mann-Kendall test, an estimate of the magnitude
of the trend may be required. In this regard, linear regression may be used to estimate the trend,
provided that the assumptions required for linear regression are met. Despite the noted
drawbacks to using regression for trend analysis, software for regression is readily available and
the technique is widely known. Regression also provides traditional confidence intervals and
hypothesis tests for evaluating the significance of a trend. Confidence intervals are also useful
for comparing trends of different analytes and different wells.
The assumptions concerning outliers and nondetects may preclude the use of linear regression
for estimating the magnitude of the trend. If there are outliers and/or nondetects in the data set, a
nonparametric method (the Theil-Sen trend line estimator) may be used to estimate the
magnitude of the trend.
7.7.2.6 Testing Multiple Wells for Trends
The Mann-Kendall test is useful for analyzing the trend in data from a single well. If the data
were collected systematically across the site at approximately the same sampling times, the
Mann-Kendall test statistics Sk for all wells may be combined to make an overall summary for
the entire set of wells. In this approach, the statistics Sk are used as a summary measure of the
trend in each well. There must be consistency in the data series across wells to make a
determination of trend that is valid across all wells.
A single statement applicable to trends across all wells is valid if the wells exhibit approximately
steady trends in the same direction (upward or downward), with roughly comparable slopes.
Formal statistical tests for the comparability of the data series across wells and for a common
trend are described in EPA QA/G9S (EPA 2006a) in the text below. Both tests are based on the
chi-squared distribution. The two tests are designed to be implemented sequentially, first testing
for comparability of slopes, then for a significant common trend across wells.
The hypothesis tests described in EPA 2006a are:
Comparability of stations. HQ: Similar dynamics affect all K stations vs. HA: At
least two stations exhibit different dynamics.
Testing for overall monotonic trend. HO*: Contaminant levels do not change
over time vs. HA *: There is an increasing (or decreasing) trend consistent across
all stations.
Draft Technical Report 198 Revised Draft - November 26, 2012
-------
Therefore, the analyst must first test for homogeneity of stations, and then, if
homogeneity is confirmed, test for an overall monotonic trend. Directions for the
test are contained in Box 4-11 and ideally, the stations in Box 4-11 should have
equal sample sizes. However, the numbers of observations at the stations can
differ slightly, because of isolated missing values, but the overall time periods
spanned must be similar. This guidance recommends that for less than 3 time
periods, an equal number of observations (a balanced design) are required. For 4
or more time periods, up to 1 missing value per sampling location may be
tolerated.
Plots of the measurements from all wells using a different symbol for each well are examined to
assess the consistency across wells. Examples of these plots are shown in Attachment D.
Detailed instructions for performing the Mann-Kendall test for multiple wells are shown in
Attachment D in Boxes D-4, D-5 and D-6.
7.7.2.7 Multiple Observations per Time Period for Multiple Wells
If multiple measurements are taken at various times and stations, then previous approaches, with
some modifications, are still applicable. Details are provided in Sections 4.3.4.2 and 4.3.4.3 of
EPA 2006a.
7.8 Analysis of Post-restoration Trends at ISR Sites
7.8.1 Trend Analysis by Well
7.8.1.1 Introduction and Examples
Figure 7-15 shows a time plot of the uranium concentrations measured in Well PR-15 at the
Crow Butte ISR site over a period of approximately 31A> years. This plot shows the measured
concentrations at 16 different times. This is one of the longest sets of measurements at a restored
ISR site available for analysis. The mean uranium concentration is also shown on the graph.
Although the mean concentration is useful for comparing the post-restoration samples with the
baseline, the mean is not essential nor relevant for evaluating trends.
Figure 7-16 shows a similar plot with modified scales. The vertical axis measures the percent
deviation of each sample from the mean. The average of the percent deviations is always at 0.
The horizontal axis measures time in the number of years since the first measurement.
Comparing the graphs in Figure 7-15 and Figure 7-16, it is clear that the linear transformations
used to modify the axes do not affect the trend. But the choice of scales does affect the units for
the slope. This choice of scales provides estimates of the slope that are easily interpreted,
expressed as a percentage of the mean per year. Using these scales, the magnitude and direction
of the trends are comparable across production units, wells, and analytes.
The blue line in Figure 7-16 is the best-fitting linear regression trend line estimated using a
statistical procedure known as least squares. The equation of the line is shown on the graph,
where y represents the vertical axis, and x represents the horizontal axis which in this case is the
time axis. The coefficient of x in this equation gives the slope of the trend line, which is equal to
Draft Technical Report 199 Revised Draft - November 26, 2012
-------
5.9784 or approximately 6%. On average, the uranium concentration in Well PR-15 is rising at
about 6% of the mean per year.
Although the symbol x is used to represent the horizontal axis in the figure, the symbol Twill be
used for the horizontal time axis in the following discussion. The linear trend equation is
expressed asyT = a + f>T , whereyr denotes the percent deviation from the mean value at time T,
and Tis the number of years from the first measurement. The percent deviation from the mean is
calculated as yT = 100- \XT — XjjX^ where Xi denotes the sample concentration at time T and
X is the average concentration in the well over the period in question. The symbol P is slope of
the line. In Figure 7-16, P = 5.9784% of the mean per year. The symbol a is called they-
intercept, which is the value of the trend line at time T=0.
The dashed red lines in Figure 7-16 show the temporal variability of the samples around the
trend line. These lines are drawn at plus and minus one standard deviation from the trend line.
The positive slope of 6% per year is small relative to the temporal variability of approximately
±60%. The question then arises whether the slope is statistically significant. To test for
significance of the slope, it is necessary to compute the standard error of the estimated slope,
written as SE($). Table 7-26 shows the regression results for uranium and radium in well PR-15,
the estimates of the slope and intercept, standard errors, R-square, ^-statistics and degrees of
freedom. The slope estimate P=5.98 is in the upper left corner of the table, and the standard error
of the slope estimate SE(fi)=10.53 is shown in the row below. The ^-statistic used to test for a
significant slope is equal to the ratio of the slope to its standard error:
tp = /3/5!£(/3) = 5.98/10.53=0.57. When the absolute value of the ^-statistic is larger than the
critical value for the regression t-test, the slope estimate is statistically significant at a specified
level of confidence. Usually a 95% level of confidence is required. A table of the critical values
for the t-test is provided in Table 7-31. The smallest critical value in the table for 14 degrees of
freedom is 1.345 for a 90% confidence level. The 7-statistic of uranium is much smaller than the
critical value; hence the positive slope observed in Figure 7-16 is not statistically significant. The
uranium concentrations are considered stable because there is no significant trend.
Table 7-26. Regression Statistics for Example in Figure 7-16
Uranium (mg/1)
Estimate->
Standard Error->
R-square->
f-statistic->
Radium (pCi/L)
Estimate->
Standard Error->
R-square->
f-statistic->
Slope Intercept
5.98 -10.07
10.53 22.85
0.02
14
0.57 -0.44
Slope Intercept
-10.06 15.27
5.24 11.63
0.32
8
-1.92 1.31
<-Degrees of Freedom
<-Degrees of Freedom
Note: An automated tool, such as the linest() function in Excel, can be used to
calculate the regression statistics.
Draft Technical Report
200
Revised Draft - November 26, 2012
-------
Figure 7-17 shows the radium concentrations in well PR-15 over the same time period. After a
sudden initial rise, the trend is down for the remaining time. The estimated slope is negative 10%
per year, indicating that radium concentrations are decreasing in this well. The standard error for
the slope in Table 7-26 is 5.24 and the ^-statistic is -1.92. In this case, the downward trend is
statistically significant at the 95% level of confidence.
One advantage of the Bayesian approach is it allows for straightforward probability comparisons
of the slope parameters. Figure 7-18 shows a Bayesian interpretation of the slope estimates
obtained using non-informative prior distributions for uranium and radium in well PR-15 at
Crow Butte. In this case, a non-informative prior distribution means a prior opinion that all
values of the slope are possible, both negative and positive, and of any magnitude. The figure
shows the posterior ^-distributions for the uranium and radium slope parameters. The uranium
posterior ^-distribution is very broad, extending well below the 0 value. Although the least
squares slope estimate for uranium is positive, the estimated slope is not statistically significant
because a large portion of the posterior distribution lies below 0. The radium posterior t-
distribution is much narrower, and almost all of the distribution lies below 0. The negative slope
estimate for radium is statistically significant.
Figure 7-19 shows the complementary cumulative distribution functions (CCDFs)25 for the
posterior distributions in Figure 7-18. These curves show the probability that the slope
(expressed as the percent of the mean per year) is greater than the value shown on the horizontal
axis. The probability that the trend in radium is greater than 0 is less than 0.05 (5%), while the
probability that the trend for uranium is greater than 0 is approximately 0.75 (75%).
Figure 7-20 shows a Bayesian comparison of the trends for radium in wells PR-8 and PR-15 at
Crow Butte. Although radium was found to have a significant downward trend in well PR-15,
radium has a significant upward trend in well PR-8.
25 The CCDF is defined as one minus the cumulative distribution function (CDF).
Draft Technical Report 201 Revised Draft - November 26, 2012
-------
1.75
| 15
1 1.25
1
0)
o ^
§1
O £ 0.75
O
E
_
'c
(0
D
0.5
0.25
0
1999
2000
2001
Date
2002
2003
• Concentration • Mean Concentration
Figure 7-15. Uranium Concentrations in Crow Butte Well PR-15
150
§ -c 100
a
50
0
D£ -50
-100
y = 5.9784x-10.066
R2 = 0.0225
0
2
Years
-Concentration Variability ±57%
•Trend Line
Figure 7-16. Deviation of Uranium Concentration from Mean with
Variability Bounds (±lo)
Draft Technical Report
202
Revised Draft - November 26, 2012
-------
y = -10.06x+ 15.275
R2 = 0.3154
0
2
Years
•Concentration -Variability -±27%
-Trend Line
Figure 7-17. Radium Concentrations in Crow Butte Well PR-15
-30
-20
-10 0 10 20 30
Slope (% of mean per year)
40
50
Radium (pCi/L) Uranium (mg/l)
Figure 7-18. Comparison of Bayesian Posterior Distributions for Regression Slope
Parameter for Uranium and Radium in Crow Butte Well PR-15
Draft Technical Report
203
Revised Draft - November 26, 2012
-------
-30
-20
-10 0 10 20 30
Slope (% of mean per year)
40
50
Radium (pCi/L) Uranium (mg/l)
Figure 7-19. Comparison of Complementary Cumulative Distribution Functions (CCDF)
for Regression Slope Parameter for Uranium and Radium in Crow Butte Well PR-15
-30
-20
-10 0 10
Slope (% of mean per year)
20
30
-Well PR-8
-Well PR-15
Figure 7-20. Comparison of Bayesian Posterior Distributions for Regression Slope
Parameter for Radium in Crow Butte Wells PR-8 and PR-15
7.8.1.2 Well-by- Well Trend Analysis at Four ISR Sites
The regression trend analysis was conducted for seven selected analytes measured in 72 wells in
six restored production units at four selected ISR sites with post-stabilization data. Table 7-27
provides a summary of the trend analysis for the seven analytes selected for the study. The
analytes include chloride, iron, pH, radium, selenium, total dissolved solids (TDS), and uranium.
Draft Technical Report
204
Revised Draft - November 26, 2012
-------
The table provides information for six regression statistics: the mean sample concentration; the
number of quarters covered by the sampling program; the number of samples collected in each
well; the slope of the trend line; the ^-statistic for the slope; and the variability around the trend
line. These results are summaries of the regressions for the number of wells shown. The table
shows the minimum, maximum, mean and standard deviation for each regression statistic.
Additional details of the trend analysis are shown in Tables F-l and F-2 in Attachment F.
Table F-l shows a summary by analyte, while Table F-2 provides details on the regressions for
each analyte in each of the six production units. Table F-3 contains the trend analysis results for
each well. This table contains the data that are summarized in Tables 7-21, F-l, and F-2. The
table identifies the production unit, analyte, well, starting date, ending date, number of tests,
R-square, variability, intercept, slope, standard error of the slope, the lower and upper bounds of
the 95% confidence interval for the slope (LCL and UCL), the ^-statistic for the slope, the ^-test
result, and the Mann-Kendall test result. The test results are recorded as a +1 if there is a
significantly positive trend, -1 if there is a significantly negative trend, and 0 if the slope is not
significant.
A summary of the estimated mean slope and variability is presented in Table 7-28. Figure 7-21
contains a bar plot of the mean slopes shown in Table 7-28. On average, radium, uranium and
iron had positive trends, while chloride and selenium showed decreasing trends. Only chloride
and uranium have a 95% confidence interval which does not include 0, indicating that only these
two mean slopes are significantly different from 0.
The mean values of the variability in Table 7-28 are plotted in Figure 7-22. Iron and selenium
had the highest variability, while chloride, TDS and pH show less variability. Radium and
uranium are in the mid-range of variability. Figure 7-23 shows the 95% confidence interval for
the mean variability of each analyte. In general, the analytes with high variability also have the
widest confidence intervals. The mean variability for iron and selenium ranges from 35% to
55%, while radium and uranium have mean variability of 20% to 35%. Figure 7-24 shows the
full range of variability encountered in the regression analysis. Although the confidence intervals
for the mean variability are all below 55%, variability as high as 100% to 200% was encountered
in individual wells.
Figure 7-25 shows a scatter plot of the standard error of the slope versus the number of samples
in the regression. Most of the wells available for analysis have only 3 to 5 samples covering a
period of approximately 1 year. The standard error of the slope estimates range over more than
three orders of magnitude for these wells. This indicates that well-by-well trend analysis is very
difficult with only 4 or 5 samples in the data set. Fewer wells were available with 7 to 10
samples, and the standard error is reduced by an order of magnitude in these wells. Only a few
wells at Crow Butte were available with more than 15 samples for uranium and radium. These
are the wells discussed in the introduction to this section of the report.
Draft Technical Report 205 Revised Draft - November 26, 2012
-------
Table 7-27. Summary of Trend Analysis at Four* ISR Sites
Statistic
Mean Concentration
Number of Quarters
Number of Tests
Slope of Trendline (%/y)
/-statistic
Variability (%)
Parameter
Chloride (mg/1)
Iron (mg/1)
pH (units)
Radium (pCi/1)
Selenium (mg/1)
TDS (mg/1)
Uranium (mg/1)
Chloride (mg/1)
Iron (mg/1)
pH (units)
Radium (pCi/1)
Selenium (mg/1)
TDS (mg/1)
Uranium (mg/1)
Chloride (mg/1)
Iron (mg/1)
pH (units)
Radium (pCi/1)
Selenium (mg/1)
TDS (mg/1)
Uranium (mg/1)
Chloride (mg/1)
Iron (mg/1)
pH (units)
Radium (pCi/1)
Selenium (mg/1)
TDS (mg/1)
Uranium (mg/1)
Chloride (mg/1)
Iron (mg/1)
pH (units)
Radium (pCi/1)
Selenium (mg/1)
TDS (mg/1)
Uranium (mg/1)
Chloride (mg/1)
Iron (mg/1)
pH (units)
Radium (pCi/1)
Selenium (mg/1)
TDS (mg/1)
Uranium (mg/1)
Number of
Wells
63
61
65
69
49
72
70
63
61
65
69
49
72
70
63
61
65
69
49
72
70
63
61
65
69
49
72
70
63
61
65
69
49
72
70
63
61
65
69
49
72
70
,„. . ,„ . ,„ Standard
Minimum Maximum Mean _ . ,.
Deviation
2.00 97.00 16.47 16.99
0.00 163.00 3.58 20.76
7.00 14.00 7.54 0.86
23.00 3422.00 404.85 513.01
0.00 0.00 0.02 0.05
280.00 1183.00 536.30 211.42
0.00 9.00 1.50 1.87
2.00 5.00 2.69 0.62
1.00 14.30 3.43 3.74
1.00 5.00 2.39 1.02
1.00 14.30 3.47 3.51
0.00 14.30 3.84 4.05
1.30 14.00 3.61 3.22
2.00 14.30 3.69 3.33
3.00 5.00 4.24 0.59
3.00 10.00 4.18 2.00
3.00 4.00 3.62 0.49
3.00 10.00 4.19 1.87
3.00 10.00 5.06 1.96
3.00 9.00 4.64 1.44
3.00 17.00 5.00 2.75
-139.27 315.36 -18.85 73.25
-422.10 600.91 10.10 189.99
-11.48 264.25 4.66 33.08
-160.01 363.03 17.50 92.69
-447.43 654.25 -12.04 187.09
-197.67 128.42 1.89 39.70
-240.54 242.26 27.39 106.59
-24.38 7.64 -0.93 4.19
-7.44 25.96 0.42 4.76
-4.87 4.43 0.14 1.39
-5.18 50.07 2.31 8.54
-4.72 4.32 -0.34 2.19
-9.96 148.78 2.51 17.84
-19.10 15.85 1.05 3.99
1.71 74.21 15.11 11.95
3.10 211.76 44.54 40.28
0.18 99.55 4.53 12.18
0.90 137.99 24.67 26.92
2.50 185.67 43.97 37.08
0.07 48.07 9.53 8.83
1.06 125.06 27.44 22.45
Christensen, MU2 & MU3 (COGEMA 2008a),
Crow Butte, MU1 (Crow Butte 2002) and Irigaray,
Highland A (Kearney 2004)
13 wells in 9 units (Irigaray
and B (Power Resources 2004)),
2004).
Draft Technical Report
206
Revised Draft - November 26, 2012
-------
Table 7-28. Mean Slope and Variability Estimates with 95% Confidence Interval for the
Mean (LCL to UCL)
Slope
Chloride (mg/1)
Iron (mg/1)
pH (units)
Radium (pCi/1)
Selenium (mg/1)
TDS (mg/1)
Uranium (mg/1)
Variability
Chloride (mg/1)
Iron (mg/1)
pH (units)
Radium (pCi/1)
Selenium (mg/1)
TDS (mg/1)
Uranium (mg/1)
LCL
-37%
-38%
-3%
-4%
-64%
-7%
2%
LCL
12%
34%
2%
18%
34%
7%
22%
Mean
-19%
10%
5%
17%
-12%
2%
27%
Mean
15%
45%
5%
25%
44%
10%
27%
UCL
-1%
58%
13%
39%
40%
11%
52%
UCL
18%
55%
7%
31%
54%
12%
33%
0
h
•Si
o> S
t|
w °
0) ^
O)
2
o
10%
-10%
-20%
-30%
10.1%
-18.9%
17.5%
4.7%
1.9%
-12.0%
27.4%
Chloride Iron (mg/l) pH (units) Radium Selenium TDS Uranium
(mg/l) (pCi/l) (mg/l) (mg/l) (mg/l)
Analyte
Figure 7-21. Slope of Trend Line Averaged over All Wells
Draft Technical Report
207
Revised Draft - November 26, 2012
-------
50%
c
8 40%
-•5- 30%
|= gi
!5 55
•2 T;
-
20%
o
10%
0%
45%
44%
15%
25%
5%
10%
Chloride Iron (mg/l) pH (units) Radium Selenium IDS Uranium
(mg/l) (pCi/l) (mg/l) (mg/l) (mg/l)
Analyte
Figure 7-22. Temporal Variability (Averaged Over All Wells)
Uranium (mg/l)
IDS (mg/l)
„. Selenium (mg/l)
+j
>
75 Radium (pCi/l)
c
<
pH (units)
Iron (mg/l)
Chloride (mg/l)
0% 10% 20% 30% 40% 50% 60%
Temporal Variability (% of mean)
Figure 7-23. 95% Confidence Interval for Mean Temporal Variability
Draft Technical Report
208
Revised Draft - November 26, 2012
-------
Uranium (mg/l)
IDS (mg/l)
Selenium (mg/l)
Radium (pCi/l)
pH (units)
Iron (mg/l)
Chloride (mg/l)
50% 100% 150% 200% 250%
Temporal Variability (% of mean)
Figure 7-24. Full Range of Temporal Variability
0
G
"C" 9
g '
LU
•§ 1
C
ra
"55
° 0
•i
A
A
j
A
A
A
A
A
A
^
I A
i
A I
A
- 1
0 5 10 15 20
Total Number of Samples
Figure 7-25. Scatter Plot of the Standard Error versus the Number of Samples
Draft Technical Report
209
Revised Draft - November 26, 2012
-------
7.8.2 Pooled Trend Analysis
The trend analyses in the previous section examined over 400 time series of samples collected
from a single well. It is also possible to use regression analysis to determine if samples collected
from a number of wells have a significant trend. Figure 7-26 shows a scatter plot of the chloride
samples collected from eight wells at Christensen Mine Unit 2. This plot uses the same scales as
in the previous section, with the percentage deviation of each sample from the well mean plotted
on the vertical axis and time on the horizontal axis. With data "pooled" from eight wells, this
plot is an example of a pooled regression trend analysis.
In a pooled analysis, data for a single analyte collected from a set of wells at all times are
included in a single regression. The regression model for the pooled trend analysis is
ytj = a + /3T, where ytj denotes the (transformed) sample value from well / at time T. Here
r=0 at the time of the earliest sample. As before, each value J/;;T is a percentage deviation from
the mean for that well, and time is expressed in years from the first sample. The model uses only
a single coefficient for the intercept, as the mean of the percentage deviations for each well is
zero,
An example of a pooled regression trend line is shown in Figure 7-26. The trend line has a
positive slope of 47% per year. Due to the larger number of samples available in pooled
regression, the standard error of the slope estimate is greatly reduced. In this example the
standard error of the slope is 11% and the ^-statistic is 3.9 with 30 degrees of freedom. The
^-statistic is larger than the critical value in Table 7-31 for 30 degrees of freedom. This confirms
that there is a significant upward trend in the chloride measurements.
The plot in Figure 7-27 shows the same data used in Figure 7-26 plotted as circles with a size
proportional to the standard error of the slope estimate obtained in the trend analysis of the eight
individual wells. The wells with the smallest circles had more precise estimates of their
individual trends than wells with larger circles. The trends for the wells with smaller circles have
trends which are near the pooled slope estimate.
Figure 7-28 shows an example of the drawbacks of relying on a linear regression model to
determine the trend. This plot shows the TDS samples in the same eight wells. The trend line has
a positive slope of 27% with a standard error of 9.9% and a ^-statistic of 2.73. Although the plot
shows that concentrations increased only in the beginning and have trended down since then, the
linear trend analysis concludes that there is a significant positive trend for TDS in these wells.
The trend in the last 3 samples is downward. In this case, it may be necessary to redo the analysis
using only the last 3 sample sets to establish the most recent trend.
The standard errors of the regressions for individual wells were shown in Figure 7-29 of the
previous section. The standard error ranged over several orders of magnitude in the analysis of
individual wells. A similar plot of the standard errors for the pooled slope estimates is shown in
Figure 7-29. The pooled sample sizes ranged from 15 to 100 depending on the production unit
and analyte. The smallest pooled sample size is approximately the same as the largest individual
well sample size shown in Figure 7-29. Using a pooled trend analysis, the standard errors are
reduced to less than 30% when the pooled sample size is 40 or larger.
Draft Technical Report 210 Revised Draft - November 26, 2012
-------
Figure 7-30 shows the results of the t-tesi for a significant slope in the pooled trend analysis. A
value of+1 on the vertical axis denotes a significantly positive slope, while a value of-1
represents a significantly negative slope. If there is no significant slope, the value is plotted at 0.
This plot shows that the likelihood of finding a significantly positive or negative test result is not
related to the pooled sample size.
A count of the significant positive and negative trends in Figure 7-30 is shown in Table 7-29. Of
the 40 pooled sets of data analyzed, 10 had a significant positive trend and 6 a significant
negative trend. The remaining 24 pooled data sets showed no significant trend. The 16 data sets
with significant trends are listed in Table 7-29. The first two rows are for chloride and TDS at
Christensen MU2, the data used in Figure 7-26 through Figure 7-28. Each of the seven analytes
was found to have a significant trend in at least one production unit. When only positive trends
are considered, 6 analytes (all but selenium) showed 1 or more positive trends.
/in
g
c on
TO zu
0
s
T£ n '
o u ,
5 i
E
o on
i: -^u ,
«i—
c
O
*•• /in
TO "^u
0)
° 60^
on
*
* ^ — "
> — •"" " * "
\ ^-*^ f
L --* *
i
* y = 42.94x- 16.43
R2 = 0.34
0.00 0.25 0.50 0.75 1.00
Years from First Sample
Figure 7-26. Christensen MU2 Chloride Samples over Time with Trend Line
Draft Technical Report
211
Revised Draft - November 26, 2012
-------
c
ra
0)
0)
E
o
c
o
+j
ra
0)
Q
-0.2 0.0 0.2 0.4 0.6 0.8
Years from First Sample
1.0
Figure 7-27. Christensen MU2 Chloride Samples over Time with Trend Line
(Size of circle is equal to the standard error of the slope estimate for each individual well; hence, samples with
smaller circles are from wells with more significant trends.)
/?n
DU
A.r\
^
TO
fl> ?r\
| 20
1
> n
1 1
S 9n *
'.P
.2
O An
fin
•
* *
*
» __ —
* _---V""~
t ^---»^ :
i——"* »
y = 27.06x- 10.36
; R2 = 0.20
0.00 0.25 0.50 0.75 1.00
Years from First Sample
Figure 7-28. Christensen MU2 TDS Samples over Time with Trend Line
Draft Technical Report
212
Revised Draft - November 26, 2012
-------
RD
g. 8°
O
V)
TR Rn
H
-------
Table 7-29. Summary of Significant Trends in Pooled Trend Analysis
Positive Trend
Negative Trend
No Trend
Total
Count
10
6
24
40
Percent
25%
15%
60%
100%
Table 7-30.
Significant Positive and Negative Trends Identified using Pooled Trend
Analysis
Unit
Christensen MU2
Christensen MU3
CrowButteMUl
Highland A
Highland B
Irigaray MU1-9
Analyte
Chloride
TDS
pH (units)
Iron
TDS
Iron
Iron
Uranium
TDS
Chloride
Uranium
Radium
TDS
Chloride
Iron
Selenium
Direction
Positive
Positive
Positive
Negative
Positive
Negative
Positive
Positive
Positive
Negative
Positive
Positive
Negative
Negative
Positive
Negative
Trend (%/yr)
43%
27%
3.8%
-124%
20%
-60%
20%
13%
2.1%
-69%
91%
51%
-13%
-57%
139%
-108%
7.9 Verify that Contaminants and Hazardous Constituent Concentrations are Below
Required Restoration Levels
This section describes traditional statistical methods for verifying that contaminants and
hazardous constituent concentrations are below required restoration levels. These methods
involve a comparison of well data collected in the baseline and after restoration is complete.
Several types of statistical hypothesis tests are recommended for conducting this comparison.
Although a geostatistical analysis is not required, these methods provide a way to convert the
baseline and post-restoration sample data into 2- and 3-D graphical representations of the
characteristics of the ore zone. For those operators with an understanding of geostatistical
software and analytical procedures, these procedures may provide better insight into the
differences between baseline characteristics and post-restoration conditions. Areas showing the
greatest differences between baseline and post-restoration conditions may be identified with
these procedures.
Geostatistical methods and other models are useful in demonstrating that the potentiometric
surface has returned to baseline conditions. This demonstration requires more than a point-by-
Draft Technical Report
214
Revised Draft - November 26, 2012
-------
point analysis. For example, at least a 2-D assessment is required to assess if the direction of the
hydraulic gradient has changed. This could be detected in multiple ways, including geostatistical
models or ModFlow analysis. Point-by-point comparisons, which do not consider the location of
the wells, are not capable of detecting a change of this type.
In this document, the hypothesis testing framework described in Section 7.3 is used to verify that
contaminants and hazardous constituent concentrations are below required restoration levels. A
hypothesis test is used to compare the post-restoration conditions to the baseline. The
comparison may be based on a statistical parameter (e.g., a mean or median) of a probability
distribution selected to best represent the population, or it may be a distribution-free comparison
of the two populations. With small sample sizes, it is difficult to demonstrate conclusively that a
particular distribution represents both populations adequately. Tests that do not assume a known
family of probability distributions (e.g., normal or lognormal) to represent the populations are
called distribution-free or nonparametric tests. A nonparametric statistical test may be more
useful for comparing two populations than one which assumes a specific distribution, because
the nonparametric tests are less sensitive to deviations from the assumed distribution.
The threshold value for the statistical test may be zero, in which case, the comparison is used to
determine whether the post-restoration well values are less than baseline levels, or the threshold
value may be a positive number representing the maximum allowable difference between the two
populations. This threshold A is defined as a "substantial difference." It is anticipated that A will
be different for each analyte.
When the baseline and post-restoration samples are not collected from the same wells, the test
involves a comparison of two independent populations. Several statistical approaches are
presented for comparing an individual well to a baseline consisting of pooled data from many
wells. Two parametric statistical methods—the two-sample Student t-test and PLs—are designed
to test for a significant difference in mean concentrations. One nonparametric test is also
presented—the WRS test.
7.9.1 Parametric Method for Determining Compliance of Individual Wells
7.9.1.1 Two-sample t-test
The two-sample t-test is a parametric test for a significant difference in the means of two data
sets, when it can be assumed that the population variances are approximately equal and the data
are approximately normally distributed or the sample sizes are large (at least 30 in both data
sets). If this is not the case, then the nonparametric WRS test procedure described below is an
alternative.
Limitations and Robustness: The two-sample t-test with equal variances performs well with
moderate violations of the assumption of normality, but not with large inequalities of variances.
The Welch-Satterthwaite version of the two-sample t-test26 is an alternative parametric method
for use if unequal variances are encountered (see Satterthwaite 1946 and Welch 1947). The t-test
' See Box 3-23, EPA QA/G9s.
Draft Technical Report
215
Revised Draft - November 26, 2012
-------
is not robust against outliers because sample means and standard deviations are sensitive to
outliers. The data should be screened for outliers using the method of Section 7.5.
Procedure for the Two-Sample t-Test (Equal Variances)
Step 1. Assume there are n compliance samples, Xi, ... Xn and TV baseline samples
Yj, ... YN. Calculate the sample means, MX and My, and the sample standard
deviations, Sx and SY of the two populations. Also compute the pooled estimate of
the standard deviation using the equation:
_ (n-l)S2x+(N-l)S2
o „ —1|
p V n+N+2
Step 2. Compute the test statistic t0 = —-—, T where A may be 0.
Step 3. Find the critical value of the ^-distribution with degrees of freedom n+N-2
and cumulative probability (l~a) for a confidence level of 100(l-a)%,
l~o- The values of Student's ^-distribution are shown in Table 7-31.
Step 4. If t0 > tn+N-2(l-a) then reject the null hypothesis that the true difference
between population means is less than A.
7.9.1.2 Prediction Limit for a Future Mean
Prediction limits (PLs) are designed to provide an upper bound for the mean of a future sample
with a specified probability equal to (1 - a), known as the confidence level of the PL. It
represents the chance (over repeated applications of the limit to many similar data sets) that the
PL will contain the mean of a future (post-operations) sample from the monitoring wells. The PL
for a future mean is similar to the parametric two-sample t-test and the nonparametric WRS test,
because the mean of the compliance samples is compared to a limit calculated using the baseline
mean. Prediction limits (PLs) assume that the future distribution and the baseline distributions
have similar shapes and differ only by a shift in concentration to a higher level. Similar
assumptions apply to the other two tests. If the baseline and future distributions have
significantly different shapes, then the Welch-Satterthwaite form of the (parametric) two-sample
^-test and the Brunner-Munzel generalization of the (nonparametric) WRS test may exhibit better
performance.
A set of TV baseline samples, pooled over all baseline wells and, if necessary, screened of outliers
using the tolerance method described in Section 7.5, is used to construct the PL. The mean of n
compliance samples from one or more wells is then compared to the PL to determine
compliance. The PL has the same mathematical form as a tolerance limit, using the mean M and
the standard deviation S of TV baseline samples:
Draft Technical Report 216 Revised Draft - November 26, 2012
-------
PL=M+KNiH(l-a)-Sif A=0 or
PL=M + A + KNin (1 - a) • S if A>0
Here KN:n(\ - a) is a multiplier that depends on the number of baseline samples TV, the number of
compliance samples «, and the desired level of confidence (1 - a).
The PL on a future mean assumes that the baseline data used to construct the limit are either
normally distributed or can be normalized by a transformation. If a transformation is used (e.g.,
the natural logarithm) and the PL is built on the transformed values, the PL should not be back-
transformed before comparing it to the compliance point data. Rather, the compliance samples
should first be transformed, and the future mean computed from the transformed compliance
measurements. Then the mean of the transformed values (e.g., log-mean) should be compared to
the PL.
Procedure for Calculating a Prediction Limit
Step 1. Calculate the sample mean M and the standard deviation S from the set of
TV baseline samples.
Step 2. If the background data are approximately normal, calculate the PL using
the equation
PL = M + A + t, N-iSA— + — where A may be 0.
\n N
The Student's lvalue used in the equations has degrees of freedom (N- 1) and the
cumulative probability (1 - a) for a confidence level of 100(1 - a)%. The values of
Student's distribution for selected values of TV- 1 are shown in Table 7-31.
Step 3. Using the PL computed in Step 2, compare the mean of the compliance
samples against the PL. If the future mean is below the PL, then the null
hypothesis of no significant difference is accepted and compliance is indicated. If
the future mean exceeds the PL, there is statistically significant evidence of an
increase in concentration over baseline levels.
Table 7-31. Critical Values of the Student's ^-Distribution
Degrees
of 90%
Freedom
(N-l) 0.1
1 3.078
2
3
4
5
6
7
.886
.638
.533
.476
.440
.415
Confidence Level =100(l-a)%
95% 97.5% 99% 99.5%
Probability of Exceeding Critical
0.05 0.025 0.01
6.314
2.920
2.353
2.132
2.015
1.943
1.895
12.706
4.303
3.182
2.776
2.571
2.447
2.365
31.821
6.965
4.541
3.747
3.365
3.143
2.998
Value (a)
0.005
63.657
9.925
5.841
4.604
4.032
3.707
3.499
99.9%
0.001
318.313
22.327
10.215
7.173
5.893
5.208
4.782
Draft Technical Report
217
Revised Draft - November 26, 2012
-------
Table 7-31. Critical Values of the Student's ^-Distribution
Degrees
of
Freedom
(N-l)
90%
Confidence Level =100(l-a)%
95% 97.5% 99% 99.5%
99.9%
Probability of Exceeding Critical Value (a)
0.1 0.05 0.025 0.01 0.005 0.001
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
.397
.383
.372
.363
.356
.350
.345
.341
.337
.333
.330
.328
.325
.323
.321
.319
.318
.316
.315
.314
.313
.311
.310
.309
.309
.308
.307
.306
.306
.305
.304
.304
.303
.303
.302
.302
.301
.301
.300
.300
.299
.299
.299
.298
.298
.298
.297
.297
.297
1.860
1.833
1.812
1.796
1.782
1.771
1.761
1.753
1.746
1.740
1.734
1.729
1.725
1.721
1.717
1.714
1.711
1.708
1.706
1.703
1.701
1.699
1.697
1.696
1.694
1.692
1.691
1.690
1.688
1.687
1.686
1.685
1.684
1.683
1.682
1.681
1.680
1.679
1.679
1.678
1.677
1.677
1.676
1.675
1.675
1.674
1.674
1.673
1.673
2.306
2.262
2.228
2.201
2.179
2.160
2.145
2.131
2.120
2.110
2.101
2.093
2.086
2.080
2.074
2.069
2.064
2.060
2.056
2.052
2.048
2.045
2.042
2.040
2.037
2.035
2.032
2.030
2.028
2.026
2.024
2.023
2.021
2.020
2.018
2.017
2.015
2.014
2.013
2.012
2.011
2.010
2.009
2.008
2.007
2.006
2.005
2.004
2.003
2.896
2.821
2.764
2.718
2.681
2.650
2.624
2.602
2.583
2.567
2.552
2.539
2.528
2.518
2.508
2.500
2.492
2.485
2.479
2.473
2.467
2.462
2.457
2.453
2.449
2.445
2.441
2.438
2.434
2.431
2.429
2.426
2.423
2.421
2.418
2.416
2.414
2.412
2.410
2.408
2.407
2.405
2.403
2.402
2.400
2.399
2.397
2.396
2.395
3.355
3.250
3.169
3.106
3.055
3.012
2.977
2.947
2.921
2.898
2.878
2.861
2.845
2.831
2.819
2.807
2.797
2.787
2.779
2.771
2.763
2.756
2.750
2.744
2.738
2.733
2.728
2.724
2.719
2.715
2.712
2.708
2.704
2.701
2.698
2.695
2.692
2.690
2.687
2.685
2.682
2.680
2.678
2.676
2.674
2.672
2.670
2.668
2.667
4.499
4.296
4.143
4.024
3.929
3.852
3.787
3.733
3.686
3.646
3.610
3.579
3.552
3.527
3.505
3.485
3.467
3.450
3.435
3.421
3.408
3.396
3.385
3.375
3.365
3.356
3.348
3.340
3.333
3.326
3.319
3.313
3.307
3.301
3.296
3.291
3.286
3.281
3.277
3.273
3.269
3.265
3.261
3.258
3.255
3.251
3.248
3.245
3.242
Draft Technical Report
218
Revised Draft - November 26, 2012
-------
Table 7-31. Critical Values of the Student's ^-Distribution
Degrees
of
Freedom
(N-l)
57
90%
Confidence Level =100(l-a)%
95% 97.5% 99% 99.5%
.297 1.672
2.002
2.394
2.665
99.9%
Probability of Exceeding Critical Value (a)
0.1 0.05 0.025 0.01 0.005 0.001
3.239
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
oo
.296
.296
.296
.296
.295
.295
.295
.295
.295
.294
.294
.294
.294
.294
.293
.293
.293
.293
.293
.293
.292
.292
.292
.292
.292
.292
.292
.292
.291
.291
.291
.291
.291
.291
.291
.291
.291
.291
.290
.290
.290
.290
.290
.282
1.672
1.671
1.671
1.670
1.670
1.669
1.669
1.669
1.668
1.668
1.668
1.667
1.667
1.667
1.666
1.666
1.666
1.665
1.665
1.665
1.665
1.664
1.664
1.664
1.664
1.663
1.663
1.663
1.663
1.663
1.662
1.662
1.662
1.662
1.662
1.661
1.661
1.661
1.661
1.661
1.661
1.660
1.660
1.645
2.002
2.001
2.000
2.000
1.999
1.998
1.998
1.997
1.997
1.996
1.995
1.995
1.994
1.994
1.993
1.993
1.993
1.992
1.992
1.991
1.991
1.990
1.990
1.990
1.989
1.989
1.989
1.988
1.988
1.988
1.987
1.987
1.987
1.986
1.986
1.986
1.986
1.985
1.985
1.985
1.984
1.984
1.984
1.960
2.392
2.391
2.390
2.389
2.388
2.387
2.386
2.385
2.384
2.383
2.382
2.382
2.381
2.380
2.379
2.379
2.378
2.377
2.376
2.376
2.375
2.374
2.374
2.373
2.373
2.372
2.372
2.371
2.370
2.370
2.369
2.369
2.368
2.368
2.368
2.367
2.367
2.366
2.366
2.365
2.365
2.365
2.364
2.326
2.663
2.662
2.660
2.659
2.657
2.656
2.655
2.654
2.652
2.651
2.650
2.649
2.648
2.647
2.646
2.645
2.644
2.643
2.642
2.641
2.640
2.640
2.639
2.638
2.637
2.636
2.636
2.635
2.634
2.634
2.633
2.632
2.632
2.631
2.630
2.630
2.629
2.629
2.628
2.627
2.627
2.626
2.626
2.576
3.237
3.234
3.232
3.229
3.227
3.225
3.223
3.220
3.218
3.216
3.214
3.213
3.211
3.209
3.207
3.206
3.204
3.202
3.201
3.199
3.198
3.197
3.195
3.194
3.193
3.191
3.190
3.189
3.188
3.187
3.185
3.184
3.183
3.182
3.181
3.180
3.179
3.178
3.177
3.176
3.175
3.175
3.174
3.090
When the baseline and post-restoration samples are not collected from the same wells, the test
involves a comparison of two independent populations.
Draft Technical Report
219
Revised Draft - November 26, 2012
-------
7.9.2 Nonparametric Tests for Comparing Baseline and Post-restoration Conditions
A comparison of post-restoration with baseline samples is conducted in Phase 4 to assess
compliance with the baseline and in Phase 5 to determine if post-restoration values have
achieved targeted remediation levels. In these comparisons, the statistical approach adopted will
depend on the type of data collected.
The statistical tests are designed to compare post-restoration parameter values with baseline well
parameters, assuming that both data sets were collected under stable conditions. It is likely that
the baseline well data will meet this condition, except for possible seasonal effects. Before
proceeding with the test for comparing baseline samples with post-restoration samples, it is first
necessary to conduct the test for homogeneity of trends and for existence of a monotonic trend as
described in Section 7.7.2 and in Attachment D in Boxes D-4, D-5 and D-6. These prior steps are
applied to the post-restoration data to affirm stability. If the test for homogeneity of trend across
wells is not met, then the individual wells should be tested for trends as described in
Section 7.7.2.1 and in Attachment D in Boxes D-l, D-2 and D-3. In this case, the following
procedures for determining if remediation goals are met are applicable only to the set of wells
with demonstrated stability.
7.9.2.1 Comparing One Well to the Baseline
A nonparametric comparison of baseline and post-restoration samples from stable wells is made
using the Wilcoxon Rank Sum (WRS) test. The WRS test also is known as the Mann-Whitney or
Wilcoxon-Mann-Whitney test. The advantage of using the nonparametric WRS test is that the
data need not have a known distribution. Given the small sample sizes, it would be difficult to
determine this distribution empirically. The WRS test also allows for nondetect measurements to
be present in the baseline and/or post-restoration samples. As a general rule, the WRS test can be
used with up to 40% "less than" measurements in either data set. Two assumptions underlying
this test are:
(1) Samples from the baseline and post-restoration periods are independent,
identically distributed random samples.
(2) Each measurement is independent of every other measurement, regardless of the
set of samples from which it came.
The null hypothesis is that the post-restoration data exceed the baseline by a substantial
difference. The null hypothesis is formulated for the express purpose of being rejected if the data
provide support for the alternative:
• The null hypothesis (Ho): The post-restoration distribution exceeds the baseline by more
than A. Symbolically, the null hypothesis is written as H0: 5 > A.
• The alternative hypothesis (HA): The post-restoration distribution does not exceed the
baseline by more than A (HA: 5 < A).
Here, A is the investigation level. The investigation level is determined on a case-by-case basis.
Draft Technical Report 220 Revised Draft - November 26, 2012
-------
The hypothesis test is structured so that the post-restoration data must provide evidence that the
site is within acceptable limits. This test assumes that any difference between the baseline and
post-restoration sample value distributions is due to a shift in the distribution of sample values to
higher values in the post-restoration period. The hypotheses to be tested using the WRS test have
the following definition.
Null Hypothesis Hn: The post—restoration distribution exceeds the baseline
distribution by more than a substantial difference delta (A);
versus the:
Alternative Hypothesis HA: The post—restoration distribution is lower than
the baseline distribution or exceeds the baseline distribution by no more than
A.
The null hypothesis is assumed to be true unless the statistical test indicates that it
should be rejected in favor of the alternative.
A two-sample statistical test examines the differences between the distributions of two
independent samples. The nw post-restoration samples from well k are compared with the N
baseline samples from all wells to determine if remediation goals have been met. The WRS test
is a test based on the relative rank of the post-restoration samples versus the baseline samples.
The WRS statistic for well k is defined as:
Wk = U k + N (N + l)/2
where £4 is equal to the number of positive differences in the set of all possible differences
between the baseline data and the post-restoration data for well k:
N «„
Here, the indicator function/[y] equals 1 ify>0 and equals 0 otherwise. Box D-7 in
Attachment D has detailed instructions for calculating the statistics C4 and Wk. For additional
information on the WRS test that is useful to nonstatisticians, see Conover 1998 (Chapter 5).
To determine if well k has met the remediation goal, the test statistic Wk is compared with the
critical value for the WRS test for sample sizes in Attachment E in Tables E-5, E-6, E-7, and E-8
for a = 0.01, 0.025, 0.05, 0.10, respectively. If the test statistic exceeds the critical value from the
table, the null hypothesis is rejected and we conclude that the parameter values in the post-
restoration period are below the baseline or exceed the baseline by no more than A.
Draft Technical Report 221 Revised Draft - November 26, 2012
-------
7.9.2.2 Comparing Multiple Wells: Testing for Homogeneity and Overall Compliance to the
Baseline
The ^-test, prediction limits and the WRS test described above are useful for analyzing the data
from a single well. After a determination of compliance is made for each well, these test results
may be combined to form an evaluation of compliance for the entire unit. The first step in this
evaluation is to plot the test results on a map of the site using one color for wells that are
determined to be in compliance and another color for the other wells. Such a plot may reveal
areas where restoration has not been successful. If geostatistical software is available, indicator
kriging may be used to estimate the percent of compliance throughout the unit. Hypothesis test
results are used to create an indicator variable for each well. The indicator variables are 1 if in
compliance and 0 if not. The resulting interpolation map would show the probabilities of
compliance in all areas of the unit. The average value of this interpolated map over the entire unit
provides a point estimate of the probability of compliance of the unit as a whole. This estimate of
compliance has an advantage that spatial autocorrelation of the wells is accounted for explicitly
in the model. However, due to the tortuous configuration of the ore zone (e,g., see Figure 7-1),
estimation of the anisotropic spatial correlation structure required for such analysis may be
difficult with the limited data available.
If geostatistical analysis is not available, an alternative approach is a statistical test for
homogeneity. To conduct a homogeneity test for compliance, the set of test statistics Wk for all
wells may be combined to make an overall summary for the entire set of wells. In this approach,
the test statistics for each well Wk are used as a summary measure of compliance in each well.
However, there must be consistency across wells in the relative levels of the baseline and post-
restoration data to make a determination of compliance that is valid across all wells.
The procedures described in Section 7.7.2 for conducting an overall test for a trend using the
summary Mann-Kendall statistics for each well may be modified to construct an overall test for
determining when remediation goals are met. Two tests are used; first a test for homogeneity
across wells and then a test for overall compliance. Again, both tests are based on the chi-
squared distribution. The two tests are designed to be implemented sequentially, testing first for
homogeneity, then for compliance across wells as follows:
Step 1. Test for comparability of wells for compliance determination
HO: Similar dynamics affect all K wells vs.
HA: At least two wells exhibit different dynamics
Step 2. Test for overall compliance
HO : Baseline values are exceeded by more than a substantial difference
A at one or more wells vs.
HA : Post-restoration values are lower than baseline values or exceed
baseline values by no more than a substantial difference A.
Draft Technical Report 222 Revised Draft - November 26, 2012
-------
When the t-test is used to test for compliance of individual wells, the test statistic for each well k
is used: zk = t0(k)as the summary comparison statistic for each well. If the WRS test is used to
determine compliance, the expected value and variance of Bunder the null distribution are:
Ek = E(Wk)= N(nn + N + l)/2
Vk = Var (Wt) = Nn „ («„ + N + l)/ 12
The standardized form of the test statistic Wk is Zk = (Wk — Ekj/^Vk . Ifzk is sufficiently large,
there is evidence that this well has met the remediation goal.
To perform the test for homogeneity (or comparability), first calculate the average of the
standardized test statistics ZM = J^Z^/K. The homogeneity chi-square statistic is/2/, = (YT^k) -
KZ2M- Using the chi-squared table in Table E-3 of Attachment E, find the critical value for/2
with (K- 1) degrees of freedom at significance level a. For example, with a significance level of
5% and 5 degrees of freedom, x\5) = 11.07; i.e., 11.07, is the cut point, which puts 5% of the
probability in the upper tail of a chi-square variable with 5 degrees of freedom. If/2/, 2%2(K-i), the wells
are not homogeneous at the significance level a. In this case, individual a -level tests should be
conducted at each well using the methods presented in Box D-7 of Attachment D.
If the hypothesis of homogeneity across wells is accepted in Step 1, use Step 2 to affirm the
compliance of all wells with the remediation goals. The chi-squared table in Table E-3 of
Attachment E is used to find the critical value for/2 with 1 degree of freedom at significance
level a*. Calculate the overall compliance test statistic/2C = KZ2M- If/2c > /2(7), reject HO* and
conclude that the site appears to be below baseline conditions or no more than A higher than
baseline conditions. If /2C < /2(/), there is not sufficient evidence (at the a significance level) that
all wells are in compliance with the remediation goals. In this case, additional remediation may
be required.
7.10 ProTJCL Software for Statistical Analysis
The ProUCL software package developed by EPA's Technology Support Center (EPA 201 Ob,
EPA 2010c) is designed to do many of the statistical tests/analyses recommended in this
document. The technical guide to Version 4.1 of the software includes the hypothesis testing
framework recommended in CERCLA guidance (EPA 2002a) and MARSSEVI (EPA 2000a).
These documents also were used to provide a framework for the hypothesis testing in this
document, and hence all use similar notation and terminology.
The package implements two of the three 2-sample tests recommended in this document for
comparison with baseline (the WRS test and two-sample t-test). Each test is implemented using
Test Form 1, Test Form 2 with A=0, and Test Form 2 with A>0. This fact makes the software
uniquely suitable for this application. The software package also includes a sample size module
for use with the parametric t-test and the WRS test. Two forms of PLs are implemented in
Version 4.1, but not the PL for the future mean of A: samples recommended for comparison with
Draft Technical Report 223 Revised Draft - November 26, 2012
-------
baseline in this document. The software package also contains a trend analysis module, including
regression analysis, the Mann-Kendall trend test and the Thiel-Sen estimate of the slope.
The package contains a variety of other parametric and nonparametric statistical methods,
including modules used for plotting the data, identifying the type of probability distribution,
parameter estimation and tolerance limits. Upper tolerance limits are used for determining if a
future observation is a part of the distribution or not. This method may be used for outlier
identification.
7.11 Summary of Statistical Approaches
The statistical approaches outlined in the previous sections are summarized here.
Phase 1 Baseline Sampling
• Estimate required number of baseline samples (Section 7.1).
• Adjust measured data for seasonality if required (Section 7.7.1 and Attachment D,
Section D.I).
• Use regression trend test or Mann-Kendall test to check for unexpected trends (Section
7.7.2 and Attachment D, Sections D.2 and D.3).
Phase 4 Establish Compliance with Baseline
• At the end of restoration, determine the number of wells to monitor (Section 7.2) and the
number of samples per well (Section 7.6) for the comparison with the baseline.
• Adjust measured individual well data for seasonality, if required (Section 7.7.1 and
Attachment D, Section D.I).
• Use the two-sample f-test (Section 7.9.1.1), PLs (Section 7.9.1.2) or the WRS test
(Section 7.9.2.1) to compare baseline to post-restoration conditions for each well or for
pooled wells (Attachment D, Section D.4).
• For multiple wells, first test wells for homogeneity. If the hypothesis of homogeneity
across all wells is accepted, then test to confirm compliance of all wells with restoration
goals. (Section 7.9.2.2 and Attachment D, Section D.5).
• If steady-state data are from different wells than the baseline data and trends are not
detected; use the two-sample t-test or the WRS test to compare baseline to steady-state
measurements for statistical differences for the pooled data of all wells combined, which
are treated as a single well. (Sections 7.9.1 and 7.9.2, and Attachment D, Section D.4).
Phase 5 Long-Term Stability Monitoring
• Determine the number of samples required to detect a trend (Section 7.7.2).
• Adjust measured data for each well for seasonality if required (Section 7.7.1 and
Attachment D, Section D.I).
Draft Technical Report 224 Revised Draft - November 26, 2012
-------
• Use the Mann-Kendall or regression trend test to test for trends for each well or in the
pooled wells (Section 7.7.2 and Attachment D, Sections D.2 and D.3).
• If trend is detected, use linear regression or Theil-Sen test to assess trend magnitude
(Section 7.7.2).
• Repeat for each well.
• If the before/after comparison is made between multiple wells, use the pooled-regression
trend test.
Gilbert 1987 contains extensive discussions of the issues concerning use of statistics in
environmental and ground water monitoring. For a detailed discussion of the tests mentioned in
this chapter, as well as step-by-step guidance on calculations for the various kinds of
comparisons, see also EPA 2000a and EPA 2006a.
Draft Technical Report 225 Revised Draft - November 26, 2012
-------
8.0 SUMMARY OF POST-CLOSURE PERFORMANCE ISSUES
This section provides a synopsis of the topics discussed in the report that are important to
designing a monitoring network and demonstrating acceptable postclosure performance of an
in-situ mining operation.
8.1 Designing the Monitoring Program to Allow Reliable Baseline Conditions to be
Established Prior to Active Mining
A meaningful interpretation of post-closure monitoring results relies on the accurate
characterization of background ground water conditions before active mining (leaching) begins.
The background (including determining baseline conditions in the production wellfield)
monitoring program must capture both temporal and spatial variability in ground water chemistry
up and down gradient from the production filed, as well as within the field. Considerations for
this performance issue include:
• Placement of monitoring wells (both within and beyond the influence of the injection-
withdrawal field) and well construction (e.g., screened intervals).
• Chemical constituents to be monitored, including sampling techniques, and frequencies.
• Duration of sampling to determine natural variations (and potential seasonality effects) in
pre-mining ground water chemistry.
• Statistical methods for assessing variations in data and confidence measures for these
data and subsequent decisions about baseline conditions (e.g., temporal variations in
"background" levels and how much data are sufficient for decision making).
The placement and number of monitoring wells in and around an in-situ mining operation are
strongly, if not totally, dependent on the site-specific hydrogeologic setting. The flow
characteristics of the ore-bearing aquifer, the injection and withdrawal rates, and the spacing of
these wells will dictate the placement of monitoring wells not only to assess baseline conditions
in the aquifer, but to enable the detection of excursions of the treated ground waters beyond the
wellfields.
Extensive experience in collecting and analyzing ground water chemical components exists
within the technical community concerned with fate and transport of pollutants. In addition,
previous investigations and restoration efforts at in-situ mining operations have produced a
substantial experience base. Sampling protocols are reasonably well developed and can be
reliably adapted to the in-situ mining application. The details of the sampling and analysis
programs are typically defined in the operating license in agreement with the appropriate
regulatory authority. We are not specifying specific requirements in detail in the Part 192
regulation, but rather deferring this responsibility to the regulatory authorities, the NRC or the
States.
The mining and post-mining restoration efforts involve actively altering the chemical
environment. Although reaction kinetics ultimately dictate how and over what time frames the
ground water chemistry will respond, the uncertainties introduced by the heterogeneities in the
Draft Technical Report 226 Revised Draft - November 26, 2012
-------
ore-bearing zone are complex and locally variable, which may limit the ability of predictive
modeling to reliably measure system responses. Statistical assessments of ground water
chemistry in monitoring well samples are the best tools for assessing the achievement of steady-
state conditions. The long-term stability (many tens of years) of the restored wellfield, however,
cannot be assured by only statistical analyses of relatively short-term (months to several years) of
post-restoration monitoring of ground water compositions in the wellfield. Geochemical
modeling of an adequately characterized rock/ground water system in the exempted aquifer
offers a more reliable projection of long-term stability to be made.
Constituents to be monitored should be established on a site-specific basis. Currently,
40 CFR Part 192 requires that molybdenum and uranium be added to the list of hazardous
constituents in 40 CFR 264.93,27 and Ra-226+Ra-228 (5 pCi/L) and gross alpha (15 pCi/L) have
been added to the concentration limits in 40 CFR 264.94.28 NRC lists these and several
additional parameters in its Standard Review Plan guidance for ISR license review (NRC 2003).
To ensure that temporal variability is captured, monitoring should be conducted over a period
long enough to capture seasonal variations. Both EPA and NRC have recommended that at least
four quarterly sets of samples be taken (40 CFR 264.97 and NRC 2003) to establish the
background levels in (i.e., the baseline) and around the production field. Since this approach
covers only one set of seasons, more samples may be required to obtain adequate statistics if
seasonal variations are anticipated. If significant seasonal variations are expected, longer time
frames for collecting samples sufficient to cover a number of seasonal cycles would be
appropriate to establish confidence in the baseline characterization. Seasonal variations are more
likely in shallow aquifers than in deeper aquifers that are more removed from recharge areas.
Ground water background measurements up gradient of the production zone could be collected
before and during the production phases to add to the database of information available to
determine the presence or absence of seasonal variations.
Monitoring for spatial variability within the permit area for mining should include wells up
gradient, down gradient, laterally adjacent to, and within the proposed leach area, sufficient to
identify zones of high and low permeability. Monitoring should also include overlying and
underlying aquifers, which could become contaminated from leaching activities. Offsite wells in
the vicinity, such as drinking water wells and stock water wells, should also be monitored. In its
Standard Review Plan for ISRs, NRC defines an acceptable set of samples as including all
wellfield perimeter monitor wells, all upper and lower aquifer monitor wells, and at least one
production/injection well per acre in each wellfield (except the requirement of one production
well per acre can be reduced for very large wellfields). It is difficult to define minimum well
spacing without detailed characterization of the flow system and injection/withdrawal rates and
configuration of the mining wellfield. Here again, we are not defining detailed requirements for
the number and placement of moniotoring wells. We emphasize that the pre-mining background
concentrations in (i.e., the baseline) and around the production field are important input
27 40 CFR 264.93 references Appendix VIII to 40 CFR Part 261 which, in turn, lists the following inorganic
species: silver, arsenic, barium, beryllium, cadmium, chromium, fluorine, mercury, nickel, lead, antimony, selenium,
thallium, and V2O5.
28 In addition to Ra-226+Ra-228 and gross alpha, 40 CFR 264.94 sets specific maximum concentrations for
arsenic, barium, cadmium, chromium, lead, mercury, selenium, and silver.
Draft Technical Report 227 Revised Draft - November 26, 2012
-------
information to support the conclusion that the wellfield ground water chemistry has been
restored to acceptable conditions. The pre-mining characterization of the ground water is also
important to support geochemical modeling of the restored preduction field and the potential fate
and transport of contaminants beyond the production field and into the down gradient portion of
the exempted aquifer. We have indicated the chemical species used for geochemical modeling in
previous sections of this report. Our listings of species to be monitored can be seen as a "tiering"
approach, in that the top tier species are those specified in regulatory language as requirements to
be met (i.e., radionuclides and toxic metals with established limits). The second tier analytes are
those needed for geochemical modeling to support decisions about the long-term stability of the
system and the behavior of radionuclides and toxic metals over the long term. This latter issue is
particularly important to support petitions for alternate concentration limits if post-restoration
activities in the production wellfield cannot reduce regulated species to acceptable levels.
8.2 Determining that the Ground Water Chemistry has Reached Steady State and
Restoration Processes can be Discontinued
Regulators must receive sufficient information so that they can determine (1) that restoration is
complete and steady-state conditions have been achieved before the initiation of post-restoration
stability monitoring, or (2) that additional restoration efforts are necessary.
As noted in EPA 1992 (Section 7.5):
Finding that the ground water has returned to a steady state after terminating
remediation efforts is an essential step in the establishment of a meaningful test of
whether or not the cleanup standards have been attained. There are uncertainties
in the process, and to some extent it is judgmental. However, if an adequate
amount of data are carefully gathered prior to beginning remediation and after
ceasing remediation, reasonable decisions can be made as to whether or not the
ground water can be considered to have reached a state of stability.
The decision on whether the ground water has reached steady state will be based
on a combination of statistical calculations, plots of data, ground water modeling,
use of predictive models, and expert advice from hydrogeologists familiar with the
site.
In addition to ground water chemistry, attention must be directed to site hydrology to establish
that the potentiometric surface has returned to approximately baseline conditions.
Restoration is expected to take several years (see Table 6-1). During this time, ground water
sampling will be used to follow the progress of the restoration process.
Considerations for this performance issue include:
• Placement of monitoring wells in and surrounding the injection-extraction field
(proximity to the extraction field), sampling frequency, and sampling techniques
(particularly if they differ from the pre-mining techniques).
Draft Technical Report 228 Revised Draft - November 26, 2012
-------
• Chemical constituents to be examined (mobilized species) and constituents that may have
been added to the ground water in attempts to restore pre-mining conditions
(e.g., chemical reducing agents or other chemicals to sequester or inhibit movement of
mobilized metals).
• Statistical tools necessary to determine when steady-state post-mining conditions are
established (data demands and consequent uncertainty levels).
The statistical tools for assessing steady-state conditions have a well-established record of
application in other contaminant remediation efforts and are easily adapted to the in-situ leaching
application. In using these tools, care must be taken to ensure that the database for the site is
detailed enough to allow clear application of any particular statistical method and interpretation
of the results. Section 7.9 describes the statistical tools that can be used to determine steady-state
conditions using both parametric and nonparametric tests. If the monitoring period is too short,
divergent data reflecting slower flow paths through the ore zone, and still active chemical
processes, could be missed, and an incorrect assessment of the aquifer's chemical state could
result. We are not mandating the use of any specific statistical method in the standards, however
the choice of statistical methods must be appropriate to the size and quality of the database
information collected, so that any regulatory decisions based upon the use of statistical measures
can be justified. We anticipate that the appropriate regulatory authority will evaluate the
defensibility of the choices made in the license based upon the quantity and quality of the
database assembled by the facility operator.
8.3 Post-restoration Stability Monitoring
After the regulators have judged that the restoration process is complete, the period of long-term
stability monitoring begins. In the past, the stability monitoring period has been set as a license
condition at about 6 months, but more recently, the period has been increased to a minimum of
1 year (Table 6-1). Field experience suggests that 1 year may not be adequate. In some cases, the
actual stability monitoring period has extended over several years to ensure that stability has
been achieved (see Attachment B). Uranium in-situ leaching locations are typically in fluvial
sandstone deposits, which exhibit lithologic heterogeneities reflecting the original depositional
environments of the deposits. The formation of the uranium deposits in these sediments also
introduces changes in the porosity and permeability of the ore zone in contrast to the surrounding
aquifer. The mining and post-mining restoration activities would further alter the local flow
regime in the ore body. In such systems, ground water flow paths through the ore body would be
anticipated to differ significantly from the surrounding media, strongly suggesting that post-
mining monitoring time frames should be longer than sometimes applied, in order to capture the
effects of locally variable flow fields.
Considerations for this performance issue include:
• Chemical constituents in pre- and post-mining waters are examined to determine if
aquifer water quality has been degraded by the leaching operations.
• Statistical measures are needed to ensure that the ground water remains stable over
several years (i.e., concentrations are not trending upward).
Draft Technical Report 229 Revised Draft - November 26, 2012
-------
• Statistical measures are needed to make decisions on whether the aquifer has achieved
restoration goals.
The statistical tools that can be used for post-restoration stability monitoring are described in
Section 7.8. Both parametric and nonparametric approaches are described to determine if a
significant trend is occurring. The procedure involves establishing DQOs for the magnitude of
the trend (%/yr) and the probability of detecting specified changes. The number of samples
required then becomes a function of the variability of the actual sequential samples about the
mean. As mentioned above, quantitative prediction of the ground water system's chemical
evolution is extremely difficult, and statistical measures to assess steady-state attainment remain
the primary tool for evaluating the success of post-mining restoration efforts.
While statistical measures can provide confidence that the ground water chemistry has stabilized
to acceptable levels within the monitoring period, these analyses do not unequivocally prove that
the system will remain in that state over much longer time frames. In the process of in-situ
mining, chemicals (oxidizing agents and complexing agents) were added to the ore zone to
moibilize the uranium and may during that process degrade the chemical mechanism that
sequestered the uranium in that location originally. The continued operation of these processes is
essential to assure that the remaining uranium in the ore zone is not mobilized by oxygenated
waters moving into the ore zone from the up gradient direction. Geochemical modeling using
field data collected before and after the restoration process within the ore zone and down
gradient from it can be used to model the chemical reducing capacity in these areas to determine
if the ground water chemical system is likely to maintain chemically reducing conditions over
the long term, and can provide added confidence to decisions about the stability of the restored
production zone.
We have proposed several options for the length of the post-restoration stability monitoring
period. The requirements of Part 192 require consistency "to the maximum extent practicable"
with RCRA regulations. In RCRA regulations, a 30-year post-closure monitoring period is
required before the license for a disposal facility can be terminated. The intent of that RCRA
monitoring period is to provide confidence that the engineered barriers will prevent the release of
contamination into the environment. We believe that the intent of ISR post-operation restoration
is fundamentally the same, preventing contamination from moving into the ground water beyond
the boundary of the exempted aquifer. We have proposed a number of alternatives that could be
used to provide confidence that the wellfield has stabilized and will remain in that condition over
the long term.
The simplest option is to establish a 30-year post-restoration monitoring period during which
continued sampling would demonstrate stability within that time frame. This represents a
significant lengthening of the stability monitoring period over past practices. Another alternative
which would offer a mechanism to shorten the period would consist of statistically
demonstrating the absence of upward trends in contaminant concentrations, and then requiring a
fixed period of additional monitoring to add a measure of confidence that the chemical system is
remaining stable. It is anticipated that this alternative would require less than the 30-year period
while still providing some confidence that the system is maintaining stability. Another
alternative to demonstrating long-term stability involves geochemical modeling to show that a
Draft Technical Report 230 Revised Draft - November 26, 2012
-------
chemically stable environment has been established and that the reducing capacity of the system
is able to maintain chemically reducing conditions over the longer term. Geochemical
contaminant fate and transport modeling can also be used to support petitions for alternative
concentration limits if restoration efforts fail to reduce some species to baseline levels. Such
modeling could be used to demonstrate that the down gradient geochemical environment in the
exempted aquifer offers the potential to reduce contaminant concentrations in the down gradient
portion of the exempted aquifer. This approach is the preferred alternative presented in the
proposed rulemaking.
Draft Technical Report 231 Revised Draft - November 26, 2012
-------
9.0 REFERENCES
Aagaard, P., and H.C. Helgeson, 1982. "Thermodynamic and kinetic constraints on reaction rates
among minerals and aqueous solutions." Am. J. Sci., 282, 237-285.
Abdesselam, A., W. Lutze, W. Gong, E.H. Nuttall, B.A. Strietelmeir, and BJ. Travis, 1999.
"Biological reduction of uranium in groundwater and subsurface soil," Science of the Total
Environment, 250 (1-3) 21-35.
Air Force, 1997. "Model Field Sampling Plan," Version 1.1. Air Force Center for Environmental
Excellence, Brooks Air Force Base, Texas.
Alpers, C.N., and D.K. Nordstrom, 1999. "Geochemical Modeling of Water-Rock Interactions in
Mining Environments." In: G.S. Plumlee and MJ. Logsdon (Eds.), The Environmental
Geochemistry of Mineral Deposits, Part A: Processes, Techniques and Health Issues, Reviews in
Economic Geology, Vol. 6A, Society of Economic Geologists, Inc., 289-324.
Allison, J.D., D.S. Brown, and K.J. Novo-Gradac, 1991.MINTEQA2/PRODEFA2, A
Geochemical assessment model for environmental systems: version 3.0 user's manual.
EPA/600/3-91/021. U.S. EPA, Environmental Research Laboratory, Athens, Georgia.
Arbogast, L., 201 la. Letter from Larry Arbogast, Uranium One Americas, Inc., to Document
Control Desk (Keith McConnell), NRC, April 19, 2011, Subject: Termination of the excursion
status for monitor well 5MW 66. NRC Agencywide Documents Access and Management System
(ADAMS) Accession Number: ML11116A144.
Arbogast, L., 201 Ib. Letter from Larry Arbogast, Uranium One Americas, Inc., to Document
Control Desk (Keith McConnell), NRC, July 8, 2011, Subject: Quarterly Progress Report of
Monitor Wells on Excursion Status - 2nd Qtr. 2011. NRC ADAMS Accession Number:
ML11196A051.
ASTM (American Society for Testing and Materials) 1992. "Standard Guide for Sampling
Groundwater Monitoring Wells." Guide D4448-85a. West Conshohcken, Pennsylvania:
American Society for Testing and Materials. 1992.
ASTM 1998. Standard Guide for Remediation of Ground Water by Natural Attenuation at
Petroleum Release Sites. Designation El943-98. West Conshohocken, Pennsylvania.
ASTM 2010. Standard Practice for Design and Installation of Groundwater Monitoring Wells.
Designation D5092. West Conshohocken, Pennsylvania.
Aziz, J.J., M. Vanderford, C.J. Newell, M. Ling, H.S. Rifai, and J.R. Gonzales, 2006.
"Monitoring and Remediation Optimization System (MAROS) Software Version 2.2 User's
Guide." Air Force Center for Environmental Excellence, Brooks City-Base, Texas.
Draft Technical Report 232 Revised Draft - November 26, 2012
-------
Ball, J.W. and D.K. Nordstrom, 1991. User's manual for WATEQ4F, with revised
thermodynamic data base and test cases for calculating speciation of major, trace, andredox
elements in natural waters. U.S. Geological Survey Open-File Report 91-183 (revised and
reprinted August 1992).
Barcelona, M.J., 1988. "Overview of the sampling process." In: Lawrence H. Keith (Ed.),
Principles of Environmental Sampling, Chaser 1. American Chemical Society Professional
Reference Book. American Chemical Society, Washington, DC, pp. 4-23, January 1988.
Barcelona, M.J., and J.A. Helfrich, 1986. "Effects of well construction materials on ground water
samples." Environmental Science and Technology 20(11): 1179-1184.
Barcelona, M.J., and J.P. Gibb, 1986. "The development of effective ground-water sampling
protocols." In: ASTM STP-963, Proceedings of the Symposium on Field Methods for Ground
Water Contamination Studies and Their Standardization for the American Society for Testing
and Materials, February 2-7, 1986, Cocoa Beach, Florida, pp. 17-26.
Barcelona, M.J., J.P. Gibb, and R.A. Miller, 1983. "A guide to the selection of materials for
monitoring well construction and ground-water sampling." EPA 600/S2-84-024, ISWS Contract
Report 327, prepared for U.S. EPA, Robert S. Kerr Environmental Research Laboratory, Ada,
Okaloma, and Environmental Monitoring Systems Laboratory, Las Vegas, Nevada.
Barcelona, M.J., J.A. Helfrich, E.E. Garske, and J.P. Gibb, 1984. "A laboratory evaluation of
ground-water sampling mechanisms." Ground Water Monitoring Review 4(2):32-41.
Barcelona, M.J., J.P. Gibb, J.A. Helfrich, and E.E. Garske, 1985a. "Practical guide for ground-
water sampling." EPA 600/S2-85-104, ISWS Contract Report 374, prepared for U.S. EPA,
Robert S. Kerr Environmental Research Laboratory, Ada, Oklahoma, and Environmental
Monitoring Systems Laboratory, Las Vegas, Nevada.
Barcelona, M.J., J.A. Helfrich, and E.E. Garske, 1985b. "Sampling tubing effects on ground
water samples." Analytical Chemistry 57(2): 460-464.
Barcelona, M.J., J.A. Helfrich, and E.E. Garske, 1986. "Field verification of sampling methods
and materials selection for ground-water contamination studies." In: ASTM-STP 963,
Proceedings of the Symposium on Field Methods for Ground Water Contamination Studies and
Their Standardization for the American Society for Testing and Materials, Cocoa Beach, Florida,
February 2-7, 1986, pp. 221-231.
Bethke, C.M., 1996. Geochemical Reaction Modeling. Oxford University Press, New York.
Bethke, C.M., 2002. The Geochemist's Workbench, Version 4.0: A User's Guide to Rxn, Act2,
Tact, React, andGtplot. University of Illinois, Urbana-Champaign, Illinois.
Bethke, C.M. and S. Yeakel, 2009. The Geochemist's Workbench®, Release 8.0, Reference
Manual. Hydrogeology Program, University of Illinois, Urbana, Illinois.
Draft Technical Report 233 Revised Draft - November 26, 2012
-------
Bickel, P. and K. Doksum, 2006. Mathematical Statistics, Basic Ideas and Selected Topics,
2nd Ed. Prentice Hall, Upper Saddle River, New Jersey.
Bowser, C.J., and B.F. Jones, 2002. "Mineralogic controls on the composition of natural waters
dominated by silicate hydrolysis." AmJSci 302:582-662.
Cameco 2012. Letter from Ken Garoutte, Cameco Resources, to John Pessehl, Wyoming
Department of Environmental Quality, dated August 17, 2012. Subject: Groundwater Restoration
Research Proposal LQD TFN 5 1/330. Cameco Resources, Permit 633. ADAMS Accession
Number: ML12235A224.
Campbell, M.J., S.A. Julious, andD.G. Altman, 1995. "Estimating sample sizes for binary,
ordered categorical, and continuous outcomes in two group comparisons," BMJ. Volume 311, 28
October 1995.
Charbonneau, M., 2009. "Current problems and recent advances in in-situ bioremediation of
uranium-contaminated sites,'" Basic Biotechnology, 5: 13-18.
Charlton, S.R., C.L. Macklin, and D.L. Parkhurst, 1997. PHREEQCI a graphical user interface
for the geochemical computer program PHREEQC. U.S. Geologic Survey Water Resources
Investigations Report 97-4222.
Charlton, S.R., and D.L. Parkhurst, 2002. PHREEQCl-a graphical user interface to the
geochemical model PHREEQC. U.S. Geologic Survey Fact Sheet FS-031-02.
Christensen, S., and J. Doherty, 2008. Predictive error dependencies when using pilot points and
singular value decomposition in groundwater model calibration. Advances in Water Resources.
31,674-700.
Church, C.D., R.T. Wilkin, C.N. Alpers, R.O. Rye, and R.B. McClesky, 2007. "Microbial sulfate
reduction and metal attenuation in pH 4 acid mine water." Geochemical Transactions.
COGEMA 1994. Wellfield Data Package, Mine Unit 4, Christensen Ranch ISL Project.
COGEMA Mining Inc. NRC ADAMS Accession Number: ML1016770445.
COGEMA 1995. Wellfield Data Package, Mine Unit 5, Christensen Ranch Project. COGEMA
Mining Inc. NRC ADAMS Accession Number: ML113120025.
COGEMA 1996. Wellfield Data Package, Mine Unit 6, Christensen Ranch Project. COGEMA
Mining Inc. NRC ADAMS Accession Number: ML101670447.
COGEMA 2005. Response to LQD/DEQ, January 10, 2005, Comments: Irigaray Wellfield
Restoration Report. TFN 4 1/170. NRC ADAMS Accession Number: ML053270037.
COGEMA 2008a. Wellfield Restoration Report, Christensen Ranch Project, Wyoming.
COGEMA Mining Inc. March 5, 2008. NRC ADAMS Accession Number: ML081060131.
Draft Technical Report 234 Revised Draft - November 26, 2012
-------
COGEMA 2008b. Restoration Data Package, Mine Unit 5, Christensen Ranch Project.
COGEMA Mining Inc. March 5, 2008. NRC ADAMS Accession Number: ML081060132.
Cohen 2008. Email from Stephan Cohen, NRC, to rlenhard@swri.org. June 19, 2008, Subject:
Documents for Crow Butte License Renewal. NRC ADAMS Accession Number: ML090580274.
Colchin, M.P., LJ. Turk, and MJ. Humenick, 1978. "Sampling of ground water baseline and
monitoring data for in-situ processes." Report #EHE-78-01, CRWR157. Texas Water Resources
Research Center, Austin, Texas.
Conover, W.J., 1998. Practical Nonparametric Statistics, 3rd Ed. John Wiley & Sons, Inc.,
Hoboken, New Jersey.
Cooley, R.L., 2004, A theory for modeling groundwater flow in heterogeneous media: U.S.
Geological Survey Professional Paper 1679, 220 p.
Crow Butte 2000. Mine Unit 1 Restoration Report, Crow Butte Uranium Project. NRC ADAMS
Accession Number: ML003677938.
Crow Butte 2002. Mine Unit 1 Groundwater Stability Data, Source Materials License SUA-
1534, Docket Number 40-8943. Letter from Michael Griffin, Crowe Butte Resources, Inc., to
Daniel Griffen, NRC. NRC ADAMS Accession Number: ML022980095.
Crow Butte 2007. SUA-1535 License Renewal Application. Crawford, Nebraska.
Crow Butte 2011. Draft License Renewal for Crow Butte, SUA-1534. Docket 40-8943.
5/23/11.NRC ADAMS Accession Number: ML110760179.
Davis, J.A., D.E. Meece, M. Kohler, and G.P. Curtis, 2004. "Approaches to surface
complexation modeling of uranium(VI) adsorption on aquifer sediments." Geochimica et
CosmochimaActa, 68:3621-3641.
Davis, J.A., G.P. Curtis, M.J. Wilkins, M. Kohler, P.M. Fox, D.L. Naftz, and J.R. Lloyd, 2006.
"Processes affecting transport of uranium in a suboxic aquifer." Physics and Chemistry of the
Earth. 31:548-555.
Demuth and Schramke 2006. Fate and Transport of Post-Restoration Groundwater Constituents
atln-Situ Uranium Leach Facilities, Petrotek Engineering Corporation & Enchemica LLC.
Prepared for Uranium Resources, Inc. May 10, 2006.
Deutsch, W.J., W.R. Martin, L.E. Eary, and R.J. Serne, 1985. Methods of Minimizing Ground-
Water Contamination from In Situ Leach Uranium Mining. NUREG/CR-3709. NRC,
Washington, DC.
DeVoto, R.H., 1978. Uranium Geology and Exploration, Colorado School of Mines, Golden,
Colorado.
Draft Technical Report 235 Revised Draft - November 26, 2012
-------
DOE (Department of Energy) 1980. An assessment report on uranium in the United States of
America: U.S. Department of Energy Report GJO-111 (80), 150 p.
Doherty, J., and R. J. Hunt, 2009. "Two easily calculated statistics for evaluating parameter
identifiability and error reduction." Journal of Hydrology. 366, 119-127
Dong, W. and S. Brooks 2008. "Formation of Aqueous MgUO2(CO3)32" Complex and Uranium
Anion Exchange onto an Exchange Resin. Environ Sci. Technol. 2008. 42, 1979-1983.
Energy Metals 2007. Application for Source Material License, Moore Ranch Uranium Project.
Technical Report: Volume II, Sections 2.9 through 10.3. NRC ADAMS Accession Number:
ML072851268.
Engesgaard, P. and K.L. Kipp, 1992. "A geochemical transport model for redox-controlled
movement of mineral fronts in groundwater flow systems: a case of nitrate removal by oxidation
of pyrite." Water Resources Research 28:2829-2843.
EPA (Environmental Protection Agency) 1989. Statistical Analysis of Ground-Water Monitoring
Data at RCRA (Resource Conservation and Recovery Act) Facilities, Interim Final Guidance.
EPA/530-SW-89-026. Washington, DC.
EPA 1992. Methods for Evaluating the Attainment of Cleanup Standards, Volume 2: Ground
Water, EPA/230/R-92/014. U.S. Environmental Protection Agency, Office of Policy, Planning,
and Evaluation, Environmental Statistics and Information Division. Washington, DC.
EPA 1999a. Use of Monitored Natural Attenuation at Superfund, RCRA Corrective Action, and
Underground Storage Tank Sites, Office of Solid Waste and Emergency Response (OSWER),
Directive 9200.4-17P, Final Draft. Washington, DC. Publication EPA/540/R-99/009. April 1999.
EPA 1999b. Understanding Variation in Partition Coefficient, Kd, Values: Volume II: Review of
Geochemistry and Available Kd Values for Cadmium, Cesium, Chromium, Lead, Plutonium,
Radon, Strontium, Thorium, Tritium (3H) and Uranium. EPA 402-R-99-004B. Office of
Radiation and Indoor Air, Office of Solid Waste and Emergency Response, EPA [and] Office of
Environmental Restoration, U.S. Department of Energy, Washington, DC. August 1999.
EPA 2000a. Multi-Agency Radiation Survey and Site Investigation Manual (MARSSIM),
Revision 1. EPA/402/R-97/016. Washington, DC.
EPA 2000b. Statistical Estimation and Visualization of Ground-water Contamination Data.
Office of Research and Development. EPA/600/R-00/034. Washington, DC.
EPA 2000c. Policy and Program Requirements for the Mandatory Agency-Wide Quality System,
EPA Order 5360.1 A2. U.S. Environmental Protection Agency. Washington, DC.
EPA 2000d. EPA Guidance for the Data Quality Objectives Process, EPA QA/G-4, EPA/600/R-
96/055. U.S. Environmental Protection Agency. Washington, DC.
Draft Technical Report 236 Revised Draft - November 26, 2012
-------
EPA 2001. EPA Requirements for Quality Assurance Project Plans, EPA QA/R-5, EPA/240/B-
01/003. U.S. Environmental Protection Agency. Washington, DC.
EPA 2002a. Guidance for Comparing Background and Chemical Concentrations in Soil for
CERCLA Sites. EPA 540-R-01-003/OSWER 9285.7-41. September 2002.
EPA 2002b. Ground Water Issue: Calculation and Use of First-Order Rate Constants for
Monitored Natural Attenuation Studies. EPA/540/S-02/500. Authors: CJ. Newell, H.S. Rifai,
J.T. Wilson, J.A. Connor, J.A. Aziz, and M.P. Suarez.
EPA 2002c. Workshop on Monitoring Oxidation-Reduction Processes for Ground Water
Restoration, EPA/600/R-02-002. Dallas, Texas.
EPA 2002d. EPA Guidance for Quality Assurance Plans, EPA QA/G-5, EPA/240/R-02/009.
U.S. Environmental Protection Agency. Washington, DC.
EPA 2002e. Guidelines for Ensuring and Maximizing the Quality, Objectivity, Utility, and
Integrity of Information Disseminated by the Environmental Protection Agency, EPA/260R-02-
008. U.S. Environmental Protection Agency. Washington, DC.
EPA 2006a. Data Quality Assessment: Statistical Methods for Practitioners. EPA QA/G-9S.
EPA/240/B-06/003. Office of Environmental Information, Washington, DC.
EPA 2006b. EPA Peer Review Handbook, 3rded., EPA/100/B-06/002. U.S. Environmental
Protection Agency. Washington, DC.
EPA 2007a. Monitored Natural Attenuation of Inorganic Contaminants in Ground Water, Vol. 1:
Technical Basis for Assessment. EPA/600/R-07/139. National Risk Management Research
Laboratory, Ground Water and Ecosystems Restoration Division. Ada, Oklahoma.
EPA 2007b. Monitored Natural Attenuation of Inorganic Contaminants in Ground Water, Vol 2:
Assessment for Non-Radionuclides Including Arsenic, Cadmium, Chromium, Copper, Lead,
Nickel, Nitrate, Perchlorate, and Selenium. EPA/600/R-07/140. National Risk Management
Research Laboratory, Ground Water and Ecosystems Restoration Division. Ada, Oklahoma.
EPA 2009. Statistical Analysis of Groundwater Monitoring Data atRCRA Facilities - Unified
Guidance Document. EPA/530/R-09/007.
EPA 2010a. Low Stress (Low Flow) Purging and Sampling Procedure for the Collection of
Groundwater Samples from Monitoring Wells. EQASOP-GW 001.
EPA 2010b. ProUCL Version 4.1 User Guide (Draft)- Statistical Software for Environmental
Applications for Data Sets with and without Nondetect Observations. EP A/600/R-07/041.
Technology Support Center, Atlanta, Georgia.
Draft Technical Report 237 Revised Draft - November 26, 2012
-------
EPA 2010c. ProUCL Version 4.1 Technical Guide (Draft)- Statistical Software for
Environmental Applications for Data Sets with andwithout Nondetect Observations.
EPA/600/R-07/041. Technology Support Center, Atlanta, Georgia.
EPA 2010d. Monitored Natural Attenuation of Inorganic Contaminants in Ground Water
Volume 3: Assessment for Radionuclides Including Tritium, Radon, Strontium, Technetium,
Uranium, Iodine, Radium, Thorium, Cesium, and Plutonium-Americium., National Risk
Management Research Laboratory (NRMRL), Cincinnati, Ohio. Publication EPA/600/R-10/093.
September 2010.
EPA 2012. Letter from W. Honker, EPA's acting director of Region 6's water quality division, to
TCEQ Executive Director Zak Covar, dated May 16, 2012.
Fanghanel, T., and V. Neck, 2002. "Aquatic chemistry and solubility phenomena of actinide
oxides/hydroxides." Lecture presented at the 10th International Symposium on Solubility
Phenomena, Varna, Bulgaria, July 22-26, 2002. Pure Appl. Chem., Vol. 74, No. 10,
pp.1895-1907.
Federal Register 2011. "Licenses, Certifications and Approvals for Materials Licenses." Vol. 76,
No. 179, pp 56951-56966. September 15, 2011.
Gammons, C.G., JJ. Metesh, and D.M. Snyder, 2005. "A survey of the geochemistry of flooded
mine shaft water in Butte, Montana." Mine Water and the Environment, 25, 100-107.
Gard, M., and J. Mahoney, 2012. Evaluating the Effects of Uranium Kd on the Restoration oflSL
Wellfields UsingPHT3D. http://www.nrc.gov/materials/uranium-recovery/public-meetings/ur-
workshop s/gard. pdf.
Garske, E.E., and M.R. Schock, 1986. "An inexpensive flow-through cell measurement system
for monitoring selected chemical parameters ground water." Ground Water Monitoring Review
6(3):79-84.
Gault, A.G., D.R. Cooke, A.T. Townsend, J.M. Charnock, and D.A. Polya, 2005. "Mechanisms
of arsenic attenuation in acid mine drainage from Mount Bischoff, western Tasmania." Science
of the Total Environment, 345, 219-228.
Gilbert, R.O., 1987. Statistical Methods for Environmental Pollution Monitoring. VanNostrand
Reinhold, New York.
Gillham, R.W., J.L. Robin, J.F. Barker, and J.A. Cherry, 1983. Ground-water monitoring and
sample bias. API Publication #4367, American Petroleum Institute, Washington, DC, June 1983.
Glynn P.O., 2003. "Modeling Np and Pu transport with a surface complexation model and
spatially variant sorption capacities: implications for reactive transport modeling and
performance assessments of nuclear waste disposal sites." Comput Geosci 29:331-349.
Draft Technical Report 238 Revised Draft - November 26, 2012
-------
Griffin 2005. Letter from Michael Griffin, Crow Butte Resources, Inc., to Gary Janosko, NRC,
June 16, 2005. Re: Source Materials License SUA-1534, Docket No. 40-8943, SM6-28 Monitor
Well Excursion. NRC ADAMS Accession Number: ML051800471.
Hall, S., 2009. Groundwater Restoration at Uranium In-Situ Recovery Mines, South Texas
Coastal Plain. U.S. Geological Survey Open-File Report 2009-1143.
Harbaugh, A.W., E.R. Banta, M.C. Hill, and M.G. McDonald, 2000. MODFLOW-2000, the U.S.
Geological Survey Modular Ground-Water Model—User guide to modularization concepts and
the ground-water flow process. U.S. Geological Survey Open-File Report 00-92.
Harris, J., J.C. Loftis, and R.H. Montgomery, 1987. "Statistical methods for characterizing
ground-water quality." Ground Water 25(2): 185-193.
Harshman, E.N., 1972. Geology and Uranium deposits, Shirley Basin Area, Wyoming. U.S.
Geological Survey Professional Paper 745.
Harshman, E.N., 1974. "Distribution of Elements in Some Roll-Type Uranium Deposits." In
Formation of Uranium Ore Deposits—Proceedings of a Symposium., Athens, May 6-10, 1974,
pp.169-183.
Hem, J.D. "Study and Interpretation of the Chemical Characteristics of Natural Water."
USGS Water Supply Paper 1473. Denver, Colorado: U.S. Geological Survey. 1970.
Hem, J.D., 1985. Study and Interpretation of the Chemical Characteristics of Natural Water,
3rd Ed. USGS Water Supply Paper 2254. U.S. Geological Survey, Reston, Virginia.
Hodges, J.L. Jr., and E.L. Lehman, 1956. "The Efficiency of some Nonparametric Competitors
of the Mest." Annals of Mathematical Statistics, 27(2):324-335.
Hou, B., 2004. Palaeochannel studies related to the Harris Greenstone Belt, Gawler Craton,
South Australia: architecture and evolution of the Kingoonya Palaeochannel System,
Report Book, 2004/1. Department of Primary Industries and Resources South Australia,
Adelaide.
Hoy, R.N., 2006. Baseline Ground Water Quality Conditions at In Situ Uranium Wellfields in
Wyoming. Paper presented at the 2006 Billings Land Reclamation Symposium, June 4-8, 2006.
Billings, Montana.
http://www.asmr.us/Publications/Conference%20Proceedings/2006%20Billings/0246-Hoy-WY-
l.pdf
HydroGeoLogic, Inc. (2005). OU-1 Annual Ground Water Monitoring Report - Former Fort
Ord, California, Appendix D: "Mann Kendall Analysis."
Ilton, E.S. et al., 2004. "Heterogeneous reduction of uranyl by micas: Crystal chemical and
solution controls." GeochimicaEt Cosmochimica Acta, 68(11): 2417-2435.
Draft Technical Report 239 Revised Draft - November 26, 2012
-------
Ilton, E.S., A. Haiduc, C.L. Cahill, A.R. Felmy, 2005. "Mica surfaces stabilize pentavalent
uranium." Inorganic Chemistry, 44(9): 2986-2988.
Ilton, E.S., N.P. Qafoku, C.X. Liu, D.A. Moore, and J.M. Zachara, 2008. "Advective removal of
intraparticle uranium from contaminated Vadose zone sediments," Hanford, US. Environmental
Science & Technology, 42(5): 1565-1571.
Irigaray 2004. Irigaray Restoration Report, Appendix B. NRC ADAMS Accession Number:
ML053270041.
Jeon, B.H., B.A. Dempsey, W.D. Burgos, M.O. Barnett, and E.E. Roden, 2005. "Chemical
reduction of U(VI) by Fe (II) at the solid-water interface using Natural and synthetic Fe(III)
oxides," Environmental Science and Technology, 39, 5642-5649.
Johnson, R.H., and M. J. Friedel, 2009. Factoring uncertainty into restoration modeling ofin-situ
leach uranium mines. U.S. Geological Survey Open-File report 2009-1024.
Johnson, R.H., M.I. Yoshino, S.M. Hall, and V.R. Shea, 2010. "Predictive Modeling Strategies
for Operations and Closure at Uranium In-Situ Recovery Mines." Mine Water and Innovative
Thinking - IMWA 2010. Sydney, Nova Scotia.
Johnson, R.H., 2011. "Reactive Transport Modeling for the Proposed Dewey Burdock Uranium
In-Situ Recovery Mine, Edgemont, South Dakota, USA." Presented at the IMWA 11th
International Mine Water Association Congress, Mine Water - Managing the Challenges,
Aachen, Germany.
Kearney 2004. Letter from W.F. Kearney, Power Resources, to Gary Janosko, NRC. RE: Smith
Ranch-Highland Uranium Project, Docket No. 408964, SUA-1548, A-Wellfield Ground Water
Restoration Information. NRC ADAMS Accession Number: ML040300369.
Koretsky, C. (2000). "The significance of surface complexation reactions in hydrologic systems:
a geochemist's perspective." 1. Hydrol. 230, 127 -171.
Langmuir, D., 1997. Aqueous Environmental Geochemistry. Prentice Hall, Upper Saddle River,
New Jersey.
Lasaga A.C., 1981. "Transition state theory," in Kinetics of GeochemicalProcesses,
A.C. Lasaga and R.J. Kirkpatrick (Eds.). Rev. in Mineralogy 8, pp. 135-69, Min. Soc. Am.
Leach 2001. Letter from M. Leach, NRC, to M. Griffin, Crow Butte Resources, Inc.,
June 26, 2001. Subject: License Amendment 11, Crow Butte Resources In Situ Leach Facility,
License No. SUA-1534.
Lee, S.Y., M.H. Baik, and J.W. Choi, 2010. "Biogenic Formation and Growth of Uraninite
(UO2)." Environ. Sci. Technol. 44, 8409-8414.
Draft Technical Report 240 Revised Draft - November 26, 2012
-------
Liggett, W.S., 1984. "Detecting elevated contamination by comparison with background." In:
G.E. Schweitzer and J.A. Santoluci (Eds.), Environmental Sampling for Hazardous Wastes.,
Chapter 13. American Society Symposium Series #267, pp. 119-128,Washington, DC.
Liggett, W.S., 1985. "Statistical aspects of designs for studying sources of contamination in
quality assurance for environmental measurements." In: J.K. Taylor and T.W. Stanley (Eds.),
ASTMSTP 867. American Society for Testing and Materials, Philadelphia, Pennsylvania,
pp. 22-40.
Liu C., J. Zachara, L. Zhong, S. Heald, Z. Wang, B. Jeon, and J.K. Fredrickson, 2009.
"Microbial reduction of intragrain U(VI) in contaminated sediment." Environmental Science and
Technology, 434928-4933.
Loftis, J.C., R.H. Montgomery, J. Harris, D. Nettles, P.S. Porter, R.C. Ward, and T.G. Sanders,
1986. "Monitoring strategies for ground-water quality management." Final report, USGS Grant
14-08-0001-G-1060, Colorado Water Resources Research Institute, Fort Collins, Colorado.
Long, P.E., S.B. Yabusaki, P.O. Meyer, C.L. Murray, and A.L. N'Guessan, 2008. Technical
Basis for Assessing Uranium Bioremediation Performance, NUREG-CR-6973, U.S. Nuclear
Regulatory Commission.
Maest, A.S., and J.R. Kuipers, 2005. "Predicting Water Quality at Hardrock Mines: Methods and
Models, Uncertainties and State-of-the-Art." Earthworks. Washington, DC.
Malapai 1988. Christensen Ranch Production Unit 3, Baseline Water Quality Data Package.
Willow Creek Satellite, Phase 1. Malapai Resources Company.
Martin, J., G Wiatzka, J. Scharer, and B. Halbert, 2005. "Case Studies that Illustrate the
Benefits, Limitations and Information Requirements of Geochemical Modelling." In
Proceedings of the 12th Annual British Columbia - MEND ARD/Ml Workshop, November 30
and December 1, Vancouver, British Columbia.
Mayer, K.U., E.O. Frind, and D.W. Blowes, 2002. "Multicomponent reactive transport modeling
in variably saturated porous media using a generalized formulation for kinetically controlled
reactions." Water Resources Research, 38, 1174, doi: 10:1029/2001WR000862.
Mayer, K.U., D.W. Blowes, and E.O. Frind, 2003. "Advances in Reactive-Transport Modeling
of Contaminant Release and Attenuation from Mine-Waste Deposits." In J.L. Jambor, D.W.
Blowes and A.I.M. Ritchie (Eds.), Environmental Aspects of Mine Wastes, Short Course Series,
Vol. 31, Mineralogical Association of Canada, 283-302.
Meyer, P.O., M. Ye, M.L. Rockhold, S.P. Neuman, and KJ. Cantrell, 2007. Combined
Estimation of Hydrogeologic Conceptual Model, Parameter, and Scenario Uncertainty with
Application to Uranium Transport at the HanfordSite 300 Area, NUREG/CR-6940, U.S.
Nuclear Regulatory Commission, Washington, DC.
Draft Technical Report 241 Revised Draft - November 26, 2012
-------
McKinley, J.P., J.M. Zachara, J. Wan, D.E. McCready, and S.M. Heald, 2007. "Geochemical
controls on contaminant uranium in vadose Hanford Formation sediments at the 200 Area and
300 Area, Hanford Site, Washington." Vadose Zone Journal, 6:1004-1017.
Misra, K.C., 2000. Understanding Mineral Deposits, 845 pp. Dordrecht, Netherlands: Kluwer
Academic.
Moncur, M.C., CJ. Ptacek, D.W. Blowes, and J.L. Jambor, 2005. "Release, transport and
attenuation of metals from an old tailings impoundment." Applied Geochemistry, 20, 639-659.
Montgomery, R.H., J.C. Loftis, and J. Harris, 1987. "Statistical characteristics of ground water
quality variables." Ground Water 25(2): 176-184.
Moon, H.S., P.R. Komolos, and P.R. Jaffe, 2009. "Biogenic U(IV) oxidation by dissolved
oxygen and nitrate in sediment after
Contaminant Hydrology, 105;18-27.
r\
oxygen and nitrate in sediment after prolonged U(VI)/Fe(III)/SO4~ reduction," Journal of
Moss, M.E., D.P. Lettenmaier, and E.F. Wood, 1978. "On the design of hydrologic data
networks." EOS-Transactions of the American Geophysical Union 59(8):772-775.
Murakami, T., T. Sato, T. Ohnuki, and H. Isobe, 2005. "Field evidence for uranium
nanocrystallization and its implications for uranium transport." Chemical Geology, 221:117-126.
Newell, C.J., S.K. Farhat, P.C. de Blanc, and J.R. Gonzales, 2002. BIOSOURGE Source
Attenuation Decision Support System and Database. Air Force Center for Environmental
Excellence, Brooks Air Force Base, Texas.
NFESC 2000. Guide to Optimal Ground water Monitoring-Interim Final. Naval Facilities
Engineering Service Center, Port Hueneme, California.
NMA (National Mining Association) 2007. Generic Environmental Report in Support of the
Nuclear Regulatory Commission's Generic Environmental Impact Statement for
In Situ Uranium Recovery Facilities.
Nordstrom, D.K., and C.N. Alpers, 1999. "Geochemistry of acid mine waters." In G.S. Plumlee,
and MJ. Logsdon (Eds.), The Environmental Geochemistry of Mineral Deposits, 6A,
pp. 133-160. Society of Economic Geologists, Littleton, Colorado.
NRC 1997. Final Environmental Impact Statement to Construct and Operate the Crownpoint
Uranium Solution Mining Project, Crownpoint, New Mexico. NUREG-1508. U.S. Nuclear
Regulatory Commission, Washington, DC.
NRC 1998. Environmental Assessment for Renewal of Source Material License No. SUA-1534 -
Crow Butte Resources Incorporated, Crow Butte Uranium Project Dawes County, Nebraska.
Docket No. 40-8943. U.S. Nuclear Regulatory Commission, Washington, DC. NRC ADAMS
Accession Number: ML071520242.
Draft Technical Report 242 Revised Draft - November 26, 2012
-------
NRC 2001. A Baseline Risk-Informed Performance-Based Approach for In-Situ Leach Uranium
Extraction Licensees. NUREG/CR-6733, U.S. Nuclear Regulatory Commission, Washington,
DC. NRC ADAMS Accession Number: ML012840152.
NRC 2003. Standard Review Plan for In Situ Leach Uranium Extraction License Applications.
Final Report. NUREG-1569. Prepared by J. Lusher. U.S. Nuclear Regulatory Commission,
Washington, DC. NRC ADAMS Accession Number: ML032250177.
NRC 2006. Groundwater Quality Monitoring Issues at In Situ Leach (ISL), Author P. Michalak
Facilities; June 28, 2006. NRC ADAMS Accession Number: ML061910547.
NRC 2007. Consideration of Geochemical Issues in Groundwater Restorations at Uranium In-
Situ Leach Mining Facilitie. NUREG/CR-6870. Authors: J.A. Davis and G.P. Curtis. U.S.
Nuclear Regulatory Commission, Washington, DC. NRC ADAMS Accession Number:
ML070600405.
NRC 2009. Generic Environmental Impact Statement for In-Situ Leach Uranium Milling
Facilities. NUREG-1910, Final Report. U.S. Nuclear Regulatory Commission, Washington, DC.
NRC 2010. Environmental Impact Statement for the Moore Ranch ISR Project in Campbell
County, Wyoming. Supplement to the Generic Environmental Impact Statement for In-Situ Leach
Uranium Milling Facilities, Final Report. NUREG-1910, Supplement 1. U.S. Nuclear
Regulatory Commission, Washington, DC. NRC ADAMS Accession Number: ML102290470.
NRC 201 la. Environmental Impact Statement for the Nichols Ranch ISR Project in Campbell
and Johnson Counties, Wyoming. NUREG-1910 Supplement 2. U.S. Nuclear Regulatory
Commission, Washington, DC. January 2011. NRC ADAMS Accession Number:
ML103440120.
NRC 201 Ib. Environmental Impact Statement for the Lost Creek ISR Project in Sweetwater
County, Wyoming. NUREG-1910 Supplement 3. U.S. Nuclear Regulatory Commission,
Washington, DC. June 2011. NRC ADAMS Accession Number: ML11125A006.
NRDC (National Resources Defense Council) 2012. Nuclear Fuel's Dirty Beginnings. G.H.
Fettus and M.G. McKinzie (auth.). March 2012.
Ochs, M., J.A. Davis, M. Olin, I.E. Payne, C.J. Tweed, M.M. Askarieh, and S. Altmann, 2006.
"Use of thermodynamic sorption models to derive radionuclide Kd values for performance
assessment: selected results and recommendations of the NEA sorption project." Radiochimica
Acta, 94:779-785.
Odeh, R.E., andD.B. Owen, 1980. Tables for Normal Tolerance Limits, Sampling Plans, and
Screening. Marcel Dekker, Inc., New York.
Draft Technical Report 243 Revised Draft - November 26, 2012
-------
O'Loughlin, E.J., S.D. Kelly, R.E. Cook, R. Csencsits, and K.M. Kemner, 2003. "Reduction of
Uranium(VI) by mixed iron(II/iron(III) hydroxide (green rust): Formation of UC>2
manoparticies." Environ. Sci. Technol. 37, 721-727.
Ortiz-Bernard, I, R.T. Anderson, H.A. Vironis, and D.R. Lovely, 2004. "Resistance of solid-
phaseU(IV) to microbial reduction during in situ bioremediation of uranium-contaminated
ground water," Applied and Environmental Microbiology, 70 (12),(7558-7560).
Parkhurst, D.L., L.N. Plummer, and D.C. Thorstenson, 1982. BALANCE- a computer program
for calculating mass transfer for geochemical reactions in ground water. U.S. Geological Survey
Water Resources Investigations Report 82-14.
Parkhurst, D.L., 1995. Users guide toPHREEQC -A computer program for speciation,
reaction-path, advective-transport, and inverse geochemical calculations. U.S. Geological
Survey, Water Resources Investigations Report 95-4227.
Parkhurst, D.L., and C.A.J. Appelo, 1999. User's guide toPHREEQC (Version!) - A computer
program for speciation, batch-reaction, one-dimensional transport, and inverse geochemical
calculations. U.S. Geological Survey Water Resources Investigations Report 99-4259.
Parkhurst, D.L., K.L. Kipp, P. Engesgaard, and S.R. Charlton, 2004. PHAST—A program for
simulating ground-water flow, solute transport, and multicomponent geochemical reactions.
U.S. Geological Survey Techniques and Methods 6-A8.
Parkhurst, D.L., K.L. Kipp, and S.R. Charlton, 2010. PHAST version 2 -Aprogramfor
simulating ground water flow, solute transport, and multicomponent geochemical reactions. U.S.
Geological Survey Techniques and Methods 6-A3 5.
Payne, T.E., and P.L. Airey, 2006. "Radionuclide migration at the Koongarra uranium deposit,
Northern Australia - Lessons from the Alligator Rivers analogue project." Physics and
Chemistry of the Earth, 31:572-586.
Pearson , FJ. Junior, CJ. Noronha, and R.W. Anderson, 1983. "Mathematical modeling of the
distribution of natural 14C, 234U, and 238U in a regional ground-water system", Radiocarbon, 35:
291-300.
Perkins, E.H., Y.K. Kharaka, W.D. Gunter, and J. D. DeBraal, 1990. Chapter 9, pp. 117-127,
Chapter DOI: 10.1021/bk-1990-0416. ch009, ACS Symposium Series, Vol. 416. ISBN13:
9780841217294eISBN: 9780841212725. December 7, 1990. Copyright © 1990 American
Chemical Society.
Pitzer, K.S., 1979. "Theory - Ion interaction approach." In R.M. Pytkowitcz (Ed.), Activity
Coefficients in Electrolyte Solutions, Vol. 1, pp. 157-208. CRC Press, Inc., Boca Raton, Florida.
Plummer, L.N., and D.L. Parkhurst, 1990. "Application of the Pitzer Equations to the PHREEQE
geochemical model." In D.C. Melchior and R.L. Bassett (Eds.), Chemical modeling of aqueous
Draft Technical Report 244 Revised Draft - November 26, 2012
-------
systems II: American Chemical Society Symposium Series 416, pp. 128-137. American
Chemical Society, Washington, DC.
Plummer, L.N., B.C. Prestemon, and D.L. Parkhurst, 1994. An interactive code (NETPATH)for
modeling NET geochemical reactions along a flow PATH—Version 2.0. U.S. Geological Survey
Water Resources Investigations Report 94-4169.
Pollock, D.W., 1994. User's Guide for MODPATH/MODPATH-PLOT, Version 3: A particle
tracking post-processing package for MODFLOW, the U. S. Geological Survey finite-difference
ground-water flow model. U.S. Geological Survey Open-File Report 94-464.
Porter, K.S., and N.M. Trautman, 1984. Seasonally in ground water quality. Draft report by
Cornell University Center for Environment Research for the U.S. Environmental Protection
Agency, Environmental Modeling Systems Laboratory, Las Vegas, Nevada.
Power Resources 2004. Mine Unit B Ground Water Restoration Report, Smith Ranch - Highland
Uranium Project. NRC ADAMS Accession Number: ML091831100.
Powertech 2009. Dew ey-Bur dock Project Application for NRC Uranium Recovery License, Fall
River and Custer Counties, South Dakota Technical Report. NRC ADAMS Accession Number:
ML092870295.
Prommer, H., 2002. "A Reactive Multicomponent Transport Model for Saturated Porous Media,
User's Manual Version 1.0." Contaminated Land Assessment and Remediation Research Centre,
The University of Edinburgh, United Kingdom.
Puls, R.W. and MJ. Barcelona, 1989. Ground Water Sampling for Metals Analyses. Superfund
Ground Water Issue. EPA/540/4-89/001. March 1989.
Puls, R.W. andMJ. Barcelona, 1996. Low-Flow (Minimal Drawdown) Ground-Water Sampling
Procedure, EPA/540/S-95/504. 12 pp.
Reisinger, H.J., D.R. Burris, and J.G. Hering, 2005. "Remediating subsurface arsenic
contamination with monitored natural attenuation." Environmental Science and Technology
39: 458A-464A.
Rojas, J.L., 1989. "Introduction to in situ leaching of uranium." In-Situ Leaching of Uranium:
Technical, Environmental, and Economic Aspects. IAEA-TECDOC-492. International Atomic
Energy Agency. Vienna, Austria.
Sanders, T.G., R.C. Ward, J.C. Loftis, T.D. Steel, D.D. Adrian, and V.Yevjevich, 1983. "Design
of networks for monitoring water quality." Water Resources Publications. Littleton, Colorado.
Sass, Ronald L. 2011. Uranium Mining in Texas: Why Is It Done That Way? James A. Baker III
Institute for Public Policy, Rice University. http://bakerinstitute.org/publications/GCC-pub-
SassUraniumMining-032811 .pdf
Draft Technical Report 245 Revised Draft - November 26, 2012
-------
Sato, T., T. Murakami, N. Yanase, H. Isobe, T.E. Payne, and P.L. Airey, 1997. "Iron nodules
scavenging uranium from ground water." Environmental Science and Technology,
31:2854-2858.
Satterthwaite, F. E., 1946. "An Approximate Distribution of Estimates of Variance
Components." Biometrics Bulletin 2: 110-114, DOT: 10.2307/3002019.
SC& A 2011. Groundwater Modeling Studies at In Situ Leaching Facilities and Evaluation of
Doses And Risks to Off-Site Receptors from Contaminated Groundwater. EPA Work Assignment
2-06, Task 2. December 2011.
Schmidt, C., 1989. "Groundwater Restorations and Stabilization at the Ruth-ISL Test Site In
Wyoming, USA." In Situ Leaching of Uranium: Technical, Environmental and Economic
Aspects. IAEA-TECDOC-492. International Atomic Energy Agency. Vienna, Austria.
Sowder, A.G., P.M. Bertsch, and PJ. Morris, 2003. "Partitioning and availability of uranium and
nickel in contaminated riparian sediments." Journal of Environmental Quality, 32:885-898.
Sposito, G., and J. Coves, 1988. SOILCHEM: A computer program for the calculation of
chemical spcciation in soils. Kearney Found. Soil Sci., University of California, Riverside,
California.
SRK 2011. Powertech Uranium Corporation, NI 43-101 Preliminary Assessment Dewey
Burdock Project. SRK Consulting. Updated February 7, 2011.
http://www.sedar.com/CheckCode.do:isessionid=OOOObPJAQqWov6XP6tTk Wd ECu:-l
Staub, W.P., N.E. Hinkle, R.E. Williams, F. Anastasi, J. Oslensky, and D. Rogruss, 1986. An
Analysis of Excursions at Selected In Situ Uranium Mines in Wyoming and Texas.
NUREG/CR-3967. U.S. Nuclear Regulatory Commission. Washington, DC.
Steefel, C.I., 2009. CrunchFlow: Software for Modeling Multicomponent Reactive Flow and
Transport. User's Manual. October 12, 2009.
Strata Energy 2010. Ross ISR Project U.S. NRC License Application, Crook County Wyoming.
NRC ADAMS Accession Number: ML110130335.
TCEQ 2010. Groundwater Classification. TCEQ Regulatory Guidance, Remediation Division,
RG-366/TRPP-8, Revised March 2010.
http://www.tceq.texas.gov/publications/rg/rg-3 66_trrp_08 .html.
Teahon 2005a. Letter from Larry Teahon, Crow Butte Resources, Inc., to Gary Janosko, NRC,
November 14, 2005. Re: Source Materials License SUA-1534, Docket No. 40-8943, CM9-16
Monitor Well Excursion. NRC ADAMS Accession Number: ML053270239.
Teahon 2005b. Letter from Larry Teahon, Crow Butte Resources, Inc., to Gary Janosko, NRC,
July 15, 2005. Re: Source Materials License SUA-1534, Docket No. 40-8943, SM6-28 Monitor
Well Excursion. NRC ADAMS Accession Number: ML052220108.
Draft Technical Report 246 Revised Draft - November 26, 2012
-------
Teahon 2006a. Letter from Larry Teahon, Crow Butte Resources, Inc., to Gary Janosko, NRC,
April 7, 2006. Re: Source Materials License SUA-1534, Docket No. 40-8943, CM8-21 Monitor
Well Excursion. NRC ADAMS Accession Number: ML061220279
Teahon 2006b. Letter from Larry Teahon, Crow Butte Resources, Inc., to Gary Janosko, NRC,
October 2, 2006. Re: Source Materials License SUA-1534, Docket No. 40-8943, PR-15 Monitor
Well Excursion. NRC ADAMS Accession Number: ML06860036.
Thorstenson, D.C., and D.L. Parkhurst, 2002. Calculation of individual isotope equilibrium
constants for implementation in geochemical models. U.S. Geologic Survey Water Resource
Investigations Report 02-4172.
Thorstenson, D.C. and D.L.Parkhurst, 2004. "Calculation of individual isotope equilibrium
constants for geochemical reactions." Geochim Cosmochim Acta 68(11):2449-2465
Todd, D.K., R.M. Tinlin, K.D. Schmidt, and L.G. Everett, 1976. Monitor ing ground water
quality: monitoring methodology. EPA 600/4-76-026. U.S. Environmental Protection Agency,
Environmental Monitoring Systems Laboratory, Las Vegas, Nevada.
Total Resources, 1993. Well/Held Data Package, Unit 2 North, Christensen Ranch Project,
WDEQ Permit to Mine No. 478. NRC ADAMS Accession Number: ML101670446.
Um, W., RJ. Serne, C.F. Brown, and G.V. Last, 2007. "U(VI) adsorption on aquifer sediments at
the Hanford Site." Journal of Contaminant Hydrology 93:255-269.
URI (Uranium Resources Inc.) 2006. Fate and Transport of Post-Restoration Groundwater
Constituents at In-Situ Uranium Leach Facilities. Petrotek and Enchemica.
URI (Uranium Resources Inc.) 2010. Restoration LaRositaMine, Duval County Texas.
Uranium Resources, Inc. NRC/NMA Uranium Recovery Workshop, May 27, 2010.
Wan, J., T. Tokunaga, E. Brodie, Z. Wang, Z. Zheng, D. Herman, T. Hazen, M. Firestone, and S.
Sutton, 2005. "Reoxidation of Bioreduced Uranoum under Reducing Conditions." Environ. Sci.
Technol. 39, 6162-6169.
Welch, B. L., 1947. "The generalization of 'Student's' problem when several different population
variances are involved," Biometrika 34: 28-35.
Wichers, D., 2006. Email from Donna Wichers, COGEMA, to Ron Linton, NRC,
August 11, 2006. RE: Request: Summary Table Irigarary Mine Unit Restoration RAI response.
NRC ADAMS Accession Number: ML062400363.
Wolery, T.J., 1992. EQ3/6, A Computer Program for Geochemical Aqueous Speciation-
Solubility Calculations: Theoretical Manual, User's Guide, and Related Documentation
(Version 7.0). UCRL-MA-11066, Pt 111. Lawrence Livermore National Laboratory. Livermore,
California.
Draft Technical Report 247 Revised Draft - November 26, 2012
-------
Yabusaki, S.B., Y. Fang, and S. Waichler, 2008. "Building conceptual models of field-scale
uranium reactive transport in a dynamic vadose zone-aquifer-river system." Water Resources
Research.
Yabusaki, S.B., Y. Fang, S.R. Waichler, andP.E. Long, 2010. Processes, Properties, and
Conditions Controlling In Situ Bioremediation of Uranium in Shallow, Alluvial Aquifers.,
NUREG/CR-7014, U.S. Nuclear Regulatory Commission.
Yeskis, D. and B. Zavala, 2002. Ground-Water Sampling Guidelines for Superfund and RCRA
Project Managers. EPA 542-S-02-001. May 2002.
Yue, S., and P. Pilon, 2004. "A comparison of the power of the Mest, Mann-Kendall and
bootstrap tests for trend detection." Hydrological Sciences, 49(1).
Zheng, C., and P.P.Wang, 1999. "MT3DMS: A modular three-dimensional multispecies
transport model formulation of advection, dispersion, and chemical reactions of contaminants in
groundwater systems; documentation and user's guide." Contract Report SERDP-99-1, U.S.
Army Engineer Research and Development Center, Vicksburg, Mississippi.
Draft Technical Report 248 Revised Draft - November 26, 2012
------- |